Keywords: Global healthcare challenges, Empowering individual healthcare decisions through technology, Medical technology - Entrepreneurship and commercialization
Abstract: This workshop will address the need for disruptions in lieu of the upcoming changes, challenges, and opportunities of global healthcare. The 4 P’s (Predict, Prevent, Personalize, Participate) in combination with the quadruple aim of future health related innovations (better Outcomes, dramatically reduced cost, improved patient and clinician experience) will need to drive biomedical engineering developments.

The current way of delivering services will be disrupted with new offerings and new business models. Exponential technologies - and the characteristic 6D’s effecting the future - will be presented and their impact on development, as well as on clinical and biomedical engineering education of the future.

One of the D’s is DEMOCRATIZATION, which ensures that access to novel technologies and services will be available to everyone on this planet. In the introductory presentation (Michael Friebe) a future outlook of healthcare related technology developments, associated changes in healthcare business models and their impact on disrupting the current system will be presented. Healthcare changes and associated developments likely start from a local perspective addressing the unit clinical needs found in a particular area and can then be adapted towards a more global view.

In back to back talks the typical biomedical engineering product innovation pathways in the US (Paolo Bonato) and India (Debdoot Sheet) will be presented with a particular focus on what triggers the development process and how the translation from a R&D environment to an actual patient use is addressed.

An actual translation of unmet clinical needs with a 100x or more potential cost saving focus (Alfredo Illanes) will complete the presentation section.

This is followed by a short Innovation Game with all participants and a joint discussion on how to stimulate a future oriented development value proposition of focussing on cost reduction and democratization“ in our biomedical development efforts.

Keywords: General and theoretical informatics - Big data analytics, Health Informatics - Behavioral health informatics, Public Health Informatics - Outcomes research
Abstract: Extreme environments, such as natural (underwater, altitude, space, geographic poles, volcanoes, desert,…) or forced/man-made (extreme sports, emergency forces, armed forces) are conditions where specific physiological adaptations in the human body are triggered to maintain physiological functionality and to ensure survival. The general goal in the medical and engineering areas can be formulated as: to enhance human comfort, performance and survival in extreme environments. This Mini-Symposia will present world-leading experts in varying research fields ranging from engineering and medicine in space applications, the extremes of firefighting, the challenges for elite athletes, and the extremes of premature birth.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Environmental noise is an important social issue that directly affects the efficiency of the students. The aim of this study is to investigate how environmental noise generated in the library affects the performance at learning commons. For this study, the noise of the library was recorded and sixteen students of Tecnologico de Monterrey, were recruited. They were divided into four groups, and two collaborative activities were undertaken with and without noise. In both scenarios, the performance and the physiological reaction of students were investigated. The results showed that the students had a 4% higher performance in a quiet environment than in a noisy one, in the same way, the heart rate increased by 3.48% and the blink rate by 22.91%. Finally, the neural electrical activity was reduced by at least 3%. The findings of the present study suggest that collaborative work is difficult to undertake in noise scenarios such as learning commons, where no appropriated policies are established and followed. Cognitive performance is lower in noisy than in quiet conditions.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: The frequency content of a sound changes the perception of acoustic features, but can headphones also change auditory information interpretation? Audio devices such as headphones are factors that have not been taken into consideration at studying sound perception, specifically in acoustic therapies. In particular, alternative treatments based on psychoacoustic effects could be more effective if the frequency response of headphones is found as a determining factor. This investigation, therefore, aims to study the brain response (in terms of electroencephalographic activity) produced by listening to pink noise in three different headphone models. Furthermore, not only the immediate response is studied, but the sound habituation (after daily exposure to pink noise for 30 days) is also investigated. The investigation findings reveal that headphones with a flatter frequency response provide more accurate acoustic information to the brain, what in turn, demands a larger number and a wider variety of mental resources, even after a habituation process.

Keywords: Signal pattern classification, Data mining and big data methods - Machine learning and deep learning methods, Neural networks and support vector machines in biosignal processing and classification
Abstract: This preliminary study reports application of a neural network classifier to the processing of previously collected data on low power radiofrequency propagation through the wrist with the goal to detect osteoporotic/osteopenic conditions. The data set used includes 67 subjects (23-94 years old, 50 females, 17 males, 27 osteoporotic/osteopenic, 40 healthy). We process the entire spectrum of the propagation coefficient through the wrist from 30 kHz to 2 GHz, with 201 sampling points in total. We found that the dichotomic diagnostic test of raw non-normalized radiofrequency data performed with the trained neural network approaches 90% specificity and ~70% sensitivity. These results are obtained without inclusion of any additional clinical risk factors. They justify that the radio transmission data are usable on their own as a predictor of bone density. With the inclusion of additional clinical risk factors, both specificity and sensitivity improve to 95% and 76% respectively. Our approach correlates well with the available DXA measurements and has the potential for screening patients at risk for fragility fractures, given the ease of implementation and low costs associated with both the technique and the equipment.

Keywords: Signal pattern classification, Nonlinear dynamic analysis - Biomedical signals, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Features extracted from the surface electromyography (sEMG) signals during the speaking tasks play an essential role in sEMG based speech recognition. However, currently there are no general rules on the optimal choice of sEMG features to achieve satisfactory performance. In this study, a total of 120 electrodes were placed on the face and neck muscles to record the high-density (HD) sEMG signals when subjects spoke ten digits in English. Then ten different time-domain features were computed from the HD sEMG signals and the classification performance of the speech recognition was thoroughly compared. The contribution of each feature was examined by using three performance metrics, which include classification accuracy, sensitivity, and F1-Score. The results showed that, among all the ten different features, the features of WFL, MAV, RMS, and LOGD were considered to be superior because they achieved higher classification accuracies with high sensitivities and higher F1-Scores across subjects/trials in the sEMG-based digit recognition tasks. The findings of this study might be of great value to choose proper signal features that are fed into the classifier in sEMG-based speech recognition.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Physiological systems modeling - Signal processing in simulation, Physiological systems modeling - Multivariate signal processing
Abstract: The blood pressure (BP) cuff offers possibilities of modulating the blood flow and propagation of pressure pulse along the artery. In our previous work, we researched methods to adapt cuff modulation techniques for pulse transit time - BP calibration and for measurement of other hemodynamic indices of potential interest to critical care, such as arterial compliance. A model characterized the response of the vasculature located directly under the cuff, but assumed that no significant changes occur in the peripheral vasculature. This study has been tailored to gain insights into the response of peripheral BP and pulse transit time to cuff inflation. Invasive BP data collected downstream from the cuff demonstrates that highly dynamic processes occur in the distal arm during cuff inflation. Mean arterial pressure increases in the distal artery by up to 20 mmHg, leading to a decrease in pulse transit time of up to 20 ms. Clinical Relevance: Such significant changes need to be taken into account in order to improve non-invasive BP estimations and to enable inference of other hemodynamic parameters from vasculature response to cuff inflation. A simple model is developed in order to reproduce the observed behaviors. The lumped-parameter model reveals opportunities for cuff modulation measurements which can reveal information on parameters such as systemic resistance, distal arterial, venous compliances and artery-vein interaction.

Keywords: Causality, Directionality, Multivariate methods
Abstract: Cross-frequency coupling of neural oscillation is widespread during the complex cognitive process. Therefore, identifying cross-frequency information flow is essential for revealing neural dynamics mechanisms in the brain network. A current method based on the information theory, phase transfer entropy (PTE), has been proved its effectiveness in estimating directional coupling in several recent studies. However, there remains some limits in PTE: (1)lack of multivariable effect, (2) poor robustness, (3)curse of dimensionality in the high dimensional system. This study introduced a novel multivariate phase transfer entropy method named “MPTENUE” to solve the above issues. In MPTENUE, it considered the influence of remaining confounding variables, which guaranteed its applicability in a multivariable system. Meanwhile, a nonuniform embedding (NUE) approach for state reconstruction was adopted to eliminate the dimensional curse problem. We performed a series of numerical simulations based on the typical Hénon map model. The results proved that the MPTENUE achieved better noise robustness and effectively avoided the curse of dimension; meanwhile, the accuracy and sensitivity can reach 96.9% and 99.2%, respectively.

Keywords: Signal pattern classification, Data mining and big data methods - Biosignal classification, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: The similarity is a fundamental measure from the homology theory in bioinformatics, and the biological sequence can be classified based on it. However, such an approach has not been utilized for electroencephalography (EEG)-based emotion recognition. To this end, the sequence generated by choosing the dominant brain rhythm owning maximum instantaneous power at each 0.2 s timestamp of the EEG signal has been proposed. Then, to recognize emotional arousal and valence, the similarity measures between pairwise sequences have been performed by dynamic time warping (DTW). After evaluations, the sequence that provides the highest accuracy has been obtained. Thus, the representative channel has been found. Besides, the appropriate time segment for emotion recognition has been estimated. Those findings helpfully exclude redundant data for assessing emotion. Results from the DEAP dataset displayed that the classification accuracies between 72%–75% can be realized by applying the single-channel data with a 5 s length, which is impressive when considering fewer data sources as the primary concern. Hence, the proposed idea would open a new way that uses the similarity measures of sequences for EEG-based emotion recognition.

Keywords: Signal pattern classification
Abstract: Motion recognition based on surface electromyogram (sEMG) recorded from the forearm is attracting attention for its applicability because it easily integrates with wearable devices and has a high signal-to-noise ratio. Inter-subject variability and inadequate data availability are common problems encountered in classifiers. Transfer learning (TL) techniques can reduce the inter-subject variability; however, when the amount of data recorded from each source subject is small, the TL-combined classifier is prone to overfitting problems. In this study, we tested the accuracy of motion recognition with and without TL when the source dataset was increased up to 10 times with a time-domain data augmentation method called mixup. The performance was evaluated using an 8-class sEMG dataset containing wearable sensing data from 25~subjects. We found that mixup improved the performance of TL-combined classifiers (support vector machine and 4-layered fully connected feedforward neural network). In future work, we plan to investigate the relationship between the amount of data and sEMG-based motion recognition by comparing multiple sEMG datasets and multiple data augmentation methods.

Keywords: Signal pattern classification, Physiological systems modeling - Signal processing in physiological systems, Nonlinear dynamic analysis - Biomedical signals
Abstract: Investing long hours in a cognitively demanding activity without adequate rest has been shown to lead to a decline in cognitive capacity. For this reason, it is crucial to know the moments in which the mental performance is low, to disconnect and recover. This paper presents the design of brain signal processing pipeline using electroencephalographic (EEG) signals to detect cognitive performance drops during sessions that require low physical activity, to determine when users should pause the execution of their current task to take a rest. The developed system is adaptable to any user without requiring prior training. The evaluation considers three mental states: attention, mental fatigue and stress as the most representative; these mental states were re-referenced using the first five minutes of each recording as a calibration period, before applying a set of rules to determine cognitive performance drops. The results showed that, for sixty-two monotonous driving simulation sessions (78.5 ± 22.4 minutes), the detection time occurred at 35.3 ± 18.9 minutes in 80.6% of the sessions, and for three studying sessions (30, 20 and 30 minutes each) the detection time occurred at 11.9, 12.3 and 8.3 minutes, respectively.

Keywords: Neural networks and support vector machines in biosignal processing and classification
Abstract: Deep learning has achieved unprecedented success in sleep stage classification tasks, which starts to pave the way for potential real-world applications. However, due to its enormous size, deployment of deep neural networks is hindered by high cost at various aspects, such as computation power, storage, network bandwidth, power consumption, and hardware complexity. For further practical applications (e.g., wearable sleep monitoring devices), there is a need for simple and compact models. In this paper, we propose a lightweight model, namely LightSleepNet, for rapid sleep stage classification based on spectrograms. Our model is assembled by a much fewer number of model parameters compared to existing ones. Furthermore, we convert the raw EEG data into spectrograms to speed up the training process. We evaluate the model performance on several public sleep datasets with different characteristics. Experimental results show that our lightweight model using spectrogram as input can achieve comparable overall accuracy and Cohen’s kappa (SHHS100: 86.7%-81.3%, Sleep-EDF: 83.7%-77.5%, Sleep-EDF-v1: 88.3%-84.5%) compared to the state-of-the-art methods on experimental datasets.

Keywords: Physiological systems modeling - Signal processing in physiological systems
Abstract: Heart rate monitoring based on photoplethysmography (PPG) is a noninvasive and inexpensive way of measuring many important cardiovascular metrics such as heart rate and heart rate variability, and has been used in many wearable devices. Unfortunately, the accuracy of the measurements is compromised by motion artifacts. We propose a theoretically sound method to reduce the motion artifacts of heart rate sensed by a commercial wristband. This method is based on outlier detection and singular spectrum analysis which enables us to reduce the movement-related noise in non-stationary signals. The results suggest that this method exhibits high correspondence to the simultaneously measured heart rate using ECG. Several metrics of heart rate variability computed from cleaned data also indicate high agreement with those obtained from ECG.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Data mining and big data methods - Biosignal classification, Nonlinear dynamic analysis - Biomedical signals
Abstract: Early neonatal seizures detection is one of the most challenging issues in Neonatal Intensive Care Units. Several EEG-based Neonatal Seizure Detectors were proposed to support the clinical staff. However, less invasive and more easily interpretable methods than EEG are still missing. In this work, we investigated if Heart Rate Variability analysis and related measures as input features of supervised classifiers could be a valid support for discriminating between newborns with seizures and seizure-free ones. The proposed methods were validated on 52 subjects (33 with seizures and 19 seizure-free) of a public dataset collected at the Helsinki University Hospital. Encouraging results are achieved using a Linear Support Vector Machine, obtaining about 87% Area Under ROC Curve. This suggests that Heart Rate Variability analysis might be a non-invasive pre-screening tool to identify newborns with seizures.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: A deep learning technique based on semantic segmentation was implemented into the blood pressure detection points field.

Two models were trained and evaluated in terms of a reference detector.

The proposed methodology outperforms the reference detector in two of the three classic benchmarks and on signals from a public database that were modified with realistic test maneuvers and artifacts.

Both models differentiate regions with valid information and artifacts.

So far, no other delineator had shown this capacity.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Physiological systems modeling - Signal processing in simulation, Principal and independent component analysis - Blind source separation
Abstract: Antenatal fetal health monitoring primarily depends on the signal analysis of abdominal or transabdominal electrocardiogram (ECG) recordings. The noninvasive approach for obtaining fetal heart rate (HR) reduces risks of potential infections and is convenient for the expectant mother. However, in addition to strong maternal ECG presence, undesirable signals due to body motion activity, muscle contractions, and certain bio-electric potentials degrade the diagnostic quality of obtained fetal ECG from abdominal ECG recordings. In this paper, we address this problem by proposing an improved framework for estimating fetal HR from non-invasively acquired abdominal ECG recordings. Since the most significant contamination is due to maternal ECG, in the proposed framework, we rely on neural network autoencoder for reconstructing maternal ECG. The autoencoder endeavors to establish the non-linear mapping between abdominal ECG and maternal ECG thus preserving inherent fetal ECG artifacts. The framework is supplemented with an existing blind-source separation (BSS) algorithm for post-treatment of residual signals obtained after subtracting reconstructed maternal ECG from abdominal ECG. Furthermore, experimental assessments on clinically-acquired subjects' recordings advocate the effectiveness of the proposed framework in comparison with conventional techniques for maternal ECG removal.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Photoplethysmogram (PPG) is a noninvasive circulatory signal related to the blood volume in tissue. PPG has an advantage in performing measurements on various sites with only one sensor. Recently, PPG has been applied to biometric authentication. If PPG-based authentication is available, it will be possible to connect the authentication and the applications seamlessly. However, "presentation attack (PA)" may be executed against the PPG-based authentication as well as other biometric authentication. To develop a PPG-based authentication system with countermeasures, we investigate a PA against the PPG-based authentication. The PA utilizes the advantage of PPG in performing measurements on various sites on each subject's body. It records PPG on non-genuine measurement site stealthily, generates a signal for spoofing based on the recorded PPG by utilizing the transfer function between multiple PPGs, and transmit the signal to the authentication device. In order to investigate the feasibility of the PA, we developed PPG-based authentication system comprising a PPG sensing device and an authentication algorithm. The sensing device includes three sensors for multiple measurement sites on a body. The authentication algorithm is based on an existing PPG-based identification algorithm. We recorded PPGs on the subjects' bodies using the developed system, and investigated the feasibility of the PA by inputting the feature values extracted from the PPGs recorded on non-genuine measurement sites to the classifier generated by the values from the PPGs recorded on genuine measurement sites in the experiment. The results indicated that the PA can occur with the probability of more than 80 % under an ideal condition. Although the experiment was only conducted in an ideal condition of the attacker, countermeasures such as replay attack prevention, liveness detection, and obtaining unique information of the measurement site are required against it.

Keywords: Nonlinear dynamic analysis - Biomedical signals, Data mining and big data methods - Machine learning and deep learning methods, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Studying the animal models of human neuropsychiatric disorders can facilitate the understanding of mechanisms of symptoms both physiologically and genetically. Previous studies have shown that ultrasonic vocalisations (USVs) of mice might be efficient markers to distinguish the wild type group and the model of autism spectrum disorder (mASD). Nevertheless, in-depth analysis of these `silence' sounds by leveraging the power of advanced computer audition technologies (e.,g., deep learning) is limited. To this end, we propose a pilot study on using a large-scale pre-trained audio neural network to extract high-level representations from the USVs of mice for the task on detection of mASD. Experiments have shown a best result reaching an unweighted average recall of 79.2% for the binary classification task in a rigorous subject-independent scenario. To the best of our knowledge, this is the first time to analyse the sounds that cannot be heard by human beings for the detection of mASD mice. The novel findings can be significant to motivate future works with according means on studying animal models of human patients.

Keywords: Coupling and synchronization - Nonlinear coupling, Physiological systems modeling - Systems identification, Nonlinear dynamic analysis - Biomedical signals
Abstract: The cyclical and progressively decreasing dynamics of electroencephalogram (EEG) based slow-wave activity (SWA) during sleep reflects the homeostatic component of sleep-wake regulation. The dynamic changes of heart rate (HR) and heart rate variability (HRV) indices during sleep also exhibit quasi-cyclic trends that appear to correlate with SWA. This article proposes a model to characterize the relationship between SWA, HR and HRV in the polar-coordinate (r-θ) domain. Polar coordinates are particularly well-suited to model cyclic shapes with simple (linear) equations in the r-θ plane. Group-level analyses and individual-level ones of the correlations between the polar-coordinate transformations of SWA and HR reveal R2 values of 0.99 and 0.95 respectively. Given that, HR and HRV can be estimated in less obtrusive ways compared to EEG. This research offers relevant options to conveniently monitor sleep SWA.

Keywords: Physiological systems modeling - Signal processing in simulation, Physiological systems modeling - Signal processing in physiological systems, Physiological systems modeling - Multivariate signal processing
Abstract: Patients undergoing mechanical lung ventilation are at risk of lung injury. A noninvasive bedside lung monitor may benefit these patients. The Inspired Sinewave Test (IST) can measure cardio-pulmonary parameters noninvasively. We propose a lung simulation to improve the measurement of pulmonary blood flow using IST. The new method was applied to 12 pigs’ data before lung injury (control) and after lung injury (ARDS model). Results using the lung simulation shown improvements in correlation in both simulated data (R2 increased from 0.98 to 1) and pigs’ data (R2 increased from <0.001 to 0.26). Paired blood flow measurements were performed by both the IST (noninvasive) and thermodilution (invasive). In the control group, the bias of the two methods was negligible (0.02L/min), and the limit of agreement was from -1.20 to 1.18 L/min. The bias was -0.68 L/min in the ARDS group and with a broader limit of agreement (-2.49 to 1.13 L/min).

Keywords: Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: Large annotated lung sound databases are publicly available and might be used to train algorithms for diagnosis systems. However, it might be a challenge to develop a well-performing algorithm for small non-public data, which have only a few subjects and show differences in recording devices and setup. In this paper, we use transfer learning to tackle the mismatch of the recording setup. This allows us to transfer knowledge from one dataset to another dataset for crackle detection in lung sounds. In particular, a single input convolutional neural network (CNN) model is pre-trained on a source domain using ICBHI 2017, the largest publicly available database of lung sounds. We use log-mel spectrogram features of respiratory cycles of lung sounds. The pre-trained network is used to build a multi-input CNN model, which shares the same network architecture for respiratory cycles and their corresponding respiratory phases. The multi-input model is then fine-tuned on the target domain of our self-collected lung sound database for classifying crackles and normal lung sounds. Our experimental results show significant performance improvements of 9.84% (absolute) in F-score on the target domain using the multi-input CNN model and transfer learning for crackle detection.

Keywords: Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis, Time-frequency and time-scale analysis - Nonstationary analysis and modeling, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Heart rate variability (HRV) has been extensively investigated as a noninvasive marker to evaluate the functionality of the autonomic nervous system (ANS). Many studies have provided Photoplethysmography (PPG) as a surrogate for ECG signal HRV measurements. Remote PPG (rPPG) has been also investigated for pulse rate variability (PRV) estimation but in controlled conditions. We remotely extracted PRV using a smartphone camera for subjects in static and lateral motion, meanwhile their respiratory rate was set to three breathing rates in an indoor illumination environment. PRV was compared with ECG-based HRV, as ground truth. The results showed a good correspondence for rPPG-based HRV with means of SDNN and RMSSD correlation coefficient greater than 0.95 in rest and greater than 0.87 in motion. The error of mean LF/HF ratio estimated from PRV was 0.1 in rest and 0.2 in lateral motion. Moreover, a statistically significant correlation was obtained between HRV and PRV power spectra and temporal signals for all performed tasks and. The obtained results contribute to confirm that remote imaging measurement of cardiac parameters is a promising, convenient, and low-cost alternative to specialized biomedical sensors in a diversity of relevant experimental maneuver.

Keywords: Data mining and big data methods - Biosignal classification, Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: Electrocardiogram (ECG)-based identification systems have been widely studied in the literature. Usually, an ECG trace needs to be segmented according to the detected R peaks to enable feature extraction from the ECGs of duration equal to nearly one cardiac cycle. Beat averaging should also be applied to reduce the influence of inter-beat variation on the extracted features and identification accuracy. Either detecting R peaks or collecting extra heartbeats for averaging will inevitably lead to a delay in the identification process. This paper proposes a deep learning-based ECG biometric identification scheme that allows identity recognition using a random ECG segment without needing R-peak detection and beat averaging. Moreover, the problem of being vulnerable to unregistered subjects in an identification system is also addressed. Experimental results demonstrated that an identification rate of 99.1% for an identification system having 235 enrollees with an equal error rate of 8.08% was achieved.

Keywords: Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis, Time-frequency and time-scale analysis - Nonstationary analysis and modeling
Abstract: Fraction and decimal magnitude processing are crucial for mathematic achievement. Previous neuroimaging results showed that fraction and decimal processing activated both overlapping and distinct neural substrates, but temporal dissociations between fraction and decimal processing remained unknown. This event-related potential (ERP) study explored differences in neural activities between magnitude processing of fractions and decimals, by examining the notation effect (fraction vs. decimal) and distance effect (far vs. close) on early components of P1/N1, P2 and N2. Results showed that decimals elicited larger N1 and smaller P1 than fractions at the parietal region. Fractions demonstrated the significant distance effect on fronto-central P2 while decimals showed the distance effect on left anterior N2. ERP results reflect distinct processing of identification and semantic access stages between fractions and decimals. Identification is located at the visual-related region with enhanced perception acuity and identification efficiency for decimals. Semantic access activates the fronto-central region associated with elaborative magnitude manipulation for fractions, while semantic access reflects automatic phonological retrieval for decimals. Our findings disintegrate the magnitude processing of fractions and decimals from identification to magnitude processing. It reveals that temporal discrepancies between fraction and decimal magnitude processing appear as early as post-stimulus 100 ms.

Keywords: Coupling and synchronization - Coherence in biomedical signal processing
Abstract: Much of our understanding of experience-dependent plasticity originates from the level of single cells and synapses through the well-established techniques of whole-cell recording and calcium imaging. The study of cortical plasticity of neural oscillatory networks remains largely unexplored. Cross-frequency coupling has become an emerging tool to study the underlying mechanisms for synchronization and interaction between local and global processes of cortical networks. The phase of low-frequency oscillations modulates the amplitude of high-frequency oscillations through a phase-amplitude coupling. Recent studies found that gamma-band oscillations associate with critical period plasticity. The existence of such mechanisms in ocular dominance plasticity is yet to be fully demonstrated. In this study, in-vivo electrophysiological methods for recording local field potentials in the primary visual cortex (V1) of anesthetized mice are employed. Our results reveal the mechanisms of neuronal oscillatory activities for the experience-dependent plasticity of developing visual cortical circuits.

Keywords: Time-frequency and time-scale analysis - Wavelets, Time-frequency and time-scale analysis - Nonstationary analysis and modeling, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Despite prevention efforts, the prevalence of work-related upper extremity musculoskeletal disorders (WRUED) is increasing. A limit in the development of preventive interventions is the lack of devices that can measure and process sEMG signals in order to provide real-time reliable information on muscular fatigue of the upper limb in relation to the physical demands of the work. In this paper, the development and evaluation of a real-time muscle fatigue detection algorithm based on sEMG will be presented. The proposed algorithm uses the median frequency of sEMG power spectrum density (PSD) obtained with the Continuous Wavelet Transform (CWT) as an indicator of the muscle fatigue level. To extend the algorithm's efficiency to dynamic tasks, a muscle contraction detection module is added in order to remove the segments when the muscle is not contracting. To assess the algorithm's performance, eight healthy adults performed simple static and dynamic shoulder tasks using different loads. The results of the proposed time-frequency method (i.e. CWT) were first compared to those of the traditional Short Time Fourier Transform (STFT). It was shown that the CWT performs better than the STFT in both static and dynamic loading conditions. The validity of the algorithm's output as a muscle fatigue indicator was verified by comparing the output's decrease rate with different loads. As expected, the algorithm's fatigue indicator decreased faster with heavier loads. It was also shown that the initial muscle fatigue estimation output is independent of the load. Finally, we studied the proposed muscle contraction detection module's efficiency to overcome issues associated with dynamic tasks. We observed a substantial improvement of the smoothness of the fatigue indicator's evolution by using of the muscle contraction detection module.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Nonlinear dynamic analysis - Biomedical signals, Signal pattern classification
Abstract: The objective of this study was to develop a computational algorithm capable of locating artifacts and identifying epileptic seizures, which specifically implementing in clinical stereoelectroencephalography (SEEG) recordings. Based on the nonstationary nature and broadband features of SEEG signals, a comprehensive strategy combined with the complex wavelet transform (CWT) and multi-layer thresholding method was implemented for both noise reduction and seizure detection. The artifacts removal pipeline integrated edge artifact removal, discrete spectrum analysis, and peak density evaluation. For automatic seizure detection, integrated power analysis and multi-dynamic thresholding were applied. The F1-score was applied to evaluate overall performance of the algorithm. The algorithm was tested using expert-marked, double-blinded, clinical SEEG data from seven patients undergoing presurgical evaluation. This approach achieved the F1 score of 0.86 for noise reduction and 0.88 for seizure detection. This offline-approach method with minimum parameter tuning procedures and no prior information required, proved to be a feasible and solid solution for clinical SEEG data evaluation. Moreover, the algorithm can be improved with additional tuning and implemented with machine learning postprocessing pipelines.

Keywords: Data mining and big data methods - Pattern recognition, Multivariate methods, Signal pattern classification
Abstract: Modern advancements have allowed society to be at the most innovative stages of technology which involves the possibility of multimodal data collection. Dartmouth dataset is a rich dataset collected over 10 weeks from 60 participants. The dataset includes different types of data but this paper focuses on 10 different smartphone sensor data and a Patient Health Questionnaire (PHQ) 9 survey that monitors the severity of depression. This paper extracts key features from smartphone data to identify depression. A multi-view bi-clustering (MVBC) algorithm is applied to categorize homogeneous behaviour subgroups. MVBC takes multiple views of sensing data as input. The algorithm inputs three views: average, trend, and location views. MVBC categorizes the subjects to low, medium and high PHQ-9 scores. Real-world data collection may have fewer sensors, allowing for less features to be extracted. This creates a focus on prioritization of features. In this body of work, minimum redundancy maximum relevance (mRMR) is applied to the sensing features to prioritize the features that better distinguish the different groups. The resulting MVBC are compared to literature to support the categorized clusters. Decision Tree (DT) 10-fold cross validation shows that our method can classify individuals into the correct subgroups using a reduced number of features to achieve an overall accuracy of 94.7%. Achieving high accuracies with reduced features allows for focus on low power analysis and edge computing applications for long-term mental health monitoring using a smartphone.

Keywords: Connectivity, Data mining and big data methods - Pattern recognition, Coupling and synchronization - Coherence in biomedical signal processing
Abstract: Increasingly, studies have shown that changes in brain network topology accompany loss of consciousness such that the functional connectivity of the prefrontal-parietal network differs significantly in anesthetized and awake states. In this work, anesthetized and awake segments of electrocorticography were selected from two monkeys. Using phase lag index, functional connectivity matrices were built in multiple frequency bands. Quantifying topological changes in brain network through graph-theoretic properties revealed significant differences between the awake and anesthetized states. Compared to the awake state, there were distinct increases in overall and Delta prefrontal–frontal connectivity, and decreases in Alpha, Beta1 and Beta2 prefrontal–frontal connectivity during the anesthetized state, which indicate a change in the topology of the small-world network. Using functional connectivity features we achieved a satisfactory classification accuracy (93.68%). Our study demonstrates that functional connectivity features are of sufficient power to distinguish awake versus anesthetized states.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Adaptive filtering
Abstract: Onset and offset detection of electromyography (EMG) data is an important step in respiratory muscle coordination assessment. Impaired respiratory coordination can indicate breathing disorders and lung diseases. In this paper, we present an algorithm for onset and offset timing detection of real-world EMG signals from respiratory muscles, which are contaminated with electrocardiogram (ECG) artifacts. The algorithm is based on the Energy Operator signal, has a low computational cost, and includes a filtering procedure to remove ECG artifacts from EMG. Analysis of EMG signals from 2 respiratory muscles of 5 participants’ data shows high agreement between the algorithm and manual method with a mean difference between two methods of 0.0407 seconds.

Keywords: Signal pattern classification - Genetic algorithms, Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: In this work, we propose to use a deep learning framework for decoding the electroencephalogram (EEG) signals of human brain activities. More specifically, we learn an end-to-end model that recognizes natural images or motor imagery by the EEG data that is collected from the corresponding human neural activities. In order to capture the temporal information encoded in the long EEG sequences, we first employ an enhanced version of Transformer, i.e., gated Transformer, on EEG signals to learn the feature representation along a sequence of embeddings. Then a fully-connected Softmax layer is used to predict the classification results of the decoded representations. To demonstrate the effectiveness of the gated Transformer approach, we conduct experiments on the image classification task for a human brain-visual dataset and the classification task for a motor imagery dataset. The experimental results show that our method achieves new state of-the-art performance compared to multiple existing methods that are widely used for EEG classification.

Keywords: Signal pattern classification
Abstract: The effective classification for imagined speech and intended speech is of great help to the development of speech-based brain-computer interfaces (BCIs). This work distinguished imagined speech and intended speech by employing the cortical EEG signals recorded from scalp. EEG signals from eleven subjects were recorded when they produced Mandarin-Chinese monosyllables in imagined speech and intended speech, and EEG features were classified by the common spatial pattern, time-domain, frequency-domain and Riemannian manifold based methods. The classification results indicated that the Riemannian manifold based method yielded the highest classification accuracy of 85.9% among the four classification methods. Moreover, the classification accuracy with the left-only brain electrode configuration was close to that with the whole brain electrode configuration. The findings of this work have potential to extend the output commands of silent speech interfaces.

Keywords: Signal pattern classification
Abstract: This work addresses the automatic segmentation of neonatal phonocardiogram (PCG) to be used in the artificial intelligence-assisted diagnosis of abnormal heart sounds. The proposed novel algorithm has a single free parameter – the maximum heart rate. The algorithm is compared with the baseline algorithm, which was developed for adult PCG segmentation. When evaluated on a large clinical dataset of neonatal PCG with a total duration of over 7h, an F1 score of 0.94 is achieved. The main features relevant for the segmentation of neonatal PCG are identified and discussed. The algorithm is able to increase the number of cardiac cycles by a factor of 5 compared to manual segmentation, potentially allowing to improve the performance of heart abnormality detection algorithms.

Keywords: Data mining and big data methods - Inter-subject variability and personalized approaches, Time-frequency and time-scale analysis - Nonstationary analysis and modeling, Physiological systems modeling - Signal processing in physiological systems
Abstract: Fetal electrocardiography (FECG) is a promising technology for non-invasive fetal monitoring. However, due to the low amplitude and non-stationary characteristics of the FECG signal, it is difficult to extract it from maternal abdominal signals. Moreover, most FECG extraction methods are based on multiple channels, which make it difficult to achieve fetal monitoring outside the clinic. This paper proposes an efficient cluster-based method for accurate FECG extraction and fetal QRS detection only using one channel signal. We designed min-max-min group as the basis for feature extraction. The extracted features are used to distinguish the different components of the abdominal signal, and finally extract the FECG signal. To verify the effectiveness of our algorithm, we conducted experiments on a public dataset and a dataset record from the Tongji Hospital. Experimental results show that our method can achieve an accuracy rate of more than 96% which is better than other algorithms.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification, Physiological systems modeling - Multivariate signal processing
Abstract: On account of privacy preserving issue and healthcare monitoring, physiological signal biometric authentication system has gained popularity in recent years. Seismocardiogram (SCG) is now easily accessible owing to the advance of wearable sensor technology. However, SCG biometric has not been widely explored due to the challenging motion artifact removal. In this paper, we design placing the sensors at different body parts under different activities to determine the best sensor location. In addition, we develop SCG noise removal algorithm and utilize machine learning approach to perform biometric authentication tasks. We validate the proposed methods on 20 healthy young adults. The dataset contains acceleration data of sitting, standing, walking, and sitting post-exercise activities with the sensor placed at the wrists, neck, heart and sternum. We demonstrate that vertical and dorsal-ventral SCG near the heart and the sternum produce reliable SCG biometric evidenced by achieving the state-of-the-art performance. Moreover, we present the efficacy of the devised noise removal procedure in the authentication during walking motion.

Clinical relevance — A seismocardiography-based biometric authentication system can help serve privacy preserving and reveal cardiovascular functioning information in clinics.

Keywords: Physiological systems modeling - Systems identification, Physiological systems modeling - Closed loop systems
Abstract: This paper applies a kernel-based nonparametric modelling method to estimate the heart rate response during treadmill exercise and proposes a model predictive control (MPC) method to perform heart rate control for an automated treadmill system. This kernel-based method introduces a kernel regularisation term, which brings prior information to the model estimation phase. By adding this prior information, the experimental protocol can be significantly simplified and only a small amount of model training experiments are needed. The model parameters were experimentally estimated from 12 participants for the treadmill exercise with a short and practical exercise protocol. The modelling results show that the model identified using the proposed method can accurately describe the heart rate response to the treadmill exercise. Based on the identified model, an MPC controller is designed to track a predefined reference heart rate profile. An advantage is the speed and acceleration of the treadmill can be limited to within a safe range for vulnerable exercisers. The proposed controller was experimentally validated in a self-developed automated treadmill system. The tracking results indicate that the desired automatic treadmill system can regulate the participants' heart rate to follow the reference profile efficiently and safely.

Keywords: Data mining and big data methods - Pattern recognition, Data mining and big data methods - Biosignal classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Multitasking motor imagery (MI) of the unilateral upper limb is potentially more valuable in stroke rehabilitation than the current conventional MI in both hands. In this paper, a novel experimental paradigm was designed to imagine two motions of unilateral upper limb, which is hand gripping and releasing, and elbow reciprocating left and right. During this experiment, the electroencephalogram (EEG) signals were collected from 10 subjects. The time and frequency domains of the EEG signals were analyzed and visualized, indicating the presence of different Event-Related Desynchronization (ERD) or Event-Related Synchronization (ERS) for the two tasks. Then the two tasks were classified through three different EEG decoding methods, in which the optimized convolutional neural network (CNN) based on FBCNet achieved an average accuracy of 67.8%, obtaining a good recognition result. This work not only can advance the studies of MI decoding of unilateral upper limb, but also can provide a basis for better upper limb stroke rehabilitation in MI-BCI.

Keywords: Time-frequency and time-scale analysis - Nonstationary analysis and modeling, Nonlinear dynamic analysis - Biomedical signals, Signal pattern classification
Abstract: A novel low-complexity method of detecting epileptic seizures from intracranial encephalography (iEEG) signals is presented. In the proposed algorithm, coastline, energy and nonlinear energy features of iEEG signals are extracted in a patient-specific two-stage seizure detection system. The detection stage of the proposed system, which extracts two times more features than the monitoring stage, is only powered on when the monitoring stage detects a seizure occurrence. A new metric is defined to demonstrate the significance of the two-stage architecture and show the time duration over which the detection stage is activated. The new parameter is called detection stage activation ratio (DAR) and it is equal to 0.272 in this work. In addition, the proposed seizure detector outperforms other algorithms which utilize a single feature or multiple features continuously in terms of sensitivity, specificity and DAR. Therefore, it is highly suitable for seizure detector implants in which reducing the power consumption is a critical factor to increase the lifetime of the implanted battery. The algorithm is implemented on a Cyclone V FPGA and has a low dynamic power of 1 𝛍W when tested on human iEEG signals of six patients from the Bern Inselspital dataset. It reaches a perfect sensitivity of 100% tested on 120 hours of iEEG data containing 24 seizure periods of six patients.

Keywords: Signal pattern classification, Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Biosignal classification
Abstract: Knowledge regarding the site-of-collapse in the upper airway in obstructive sleep apnoea (OSA) has relevance for treatment options and their outcomes. However, current methods to identify the site-of-collapse are not suitable for clinical practice due to the invasive nature, time/expense of the tests and inconsistency of the obstruction site identified with natural and drug-induced sleep. In this study, we adopted an unsupervised algorithm to identify the predominant site-of-collapse of the upper airway during natural sleep using nocturnal audio recordings. Nocturnal audio was recorded along with full-night polysomnography with a ceiling microphone. Various acoustic features of the snore signal during hypopnoea events were extracted. We developed a feature selection algorithm combining silhouette analysis with the Laplacian score algorithm to select the high performing features. A k-means clustering model was developed to form clusters using the features extracted from snore data and analyse the correlation between the clusters generated and the predominant site-of-collapse. Cluster analysis showed that the data tends to fit well in two clusters with a mean silhouette coefficient of 0.79 and with an accuracy of 68% for classifying tongue/non-tongue collapse. The results indicate a correlation between snoring and the predominant site-of-collapse. Therefore, it could potentially be used as a practical, non-invasive, low-cost diagnosis tool for improving the selection of appropriate therapy for OSA patients without any additional burden to the patients undergoing a sleep test.

Keywords: Data mining and big data methods - Inter-subject variability and personalized approaches, Signal pattern classification, Principal component analysis
Abstract: Annotation of polysomnography (PSG) recordings for diagnosis of obstructive sleep apnea (OSA) is a standard procedure but an expensive and time-consuming process for clinicians. To aid clinicians in this process we present a data driven unsupervised hierarchical clustering approach for detection and visual presentation of breathing patterns in PSG recordings. The aim was to develop a model independent of manual annotations to detect and visualize respiratory events related to OSA. 10 recordings from the Sleep Heart Health Study database were used, and the proposed algorithm was evaluated based on the manually annotated events for each recording. The algorithm reached an F1-score of 0.58 across the 10 recordings when detecting the presence of an event vs. no event and a 100% correct diagnosis prediction of OSA when predicting if apnea-hypopnea index (AHI) ≥ 15, which is a clinically meaningful cut-off. The F1-score may be due to imprecise placement of events, difficulty distinguishing between hypopneas and stable breathing, and variations in scoring. In conclusion the performance can be improved despite the strong agreement in diagnostics. The method is a proof of concept that a clustering method can detect and visualize breathing patterns related to OSA while maintaining a correct diagnosis.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Nonlinear dynamic analysis - Biomedical signals, Signal pattern classification
Abstract: The lack of an integral characterization of chronic neuropathic pain (NP) has led to pharmacotherapy mismanagement and has hindered advances in clinical trials. In this study, we attempted to identify chronic NP by fusing psychometric (based on the Brief Inventory of Pain – BIP), and both linear and non-linear electroencephalographic (EEG) features. For this purpose, 35 chronic NP patients were recruited voluntarily. All the volunteers answered the BIP; and additionally, 22 EEG channels positioned in accordance with the 10/20 international system were registered for 10 minutes at resting state: 5 minutes with eyes open and 5 minutes with eyes closed. EEG Signals were sampled at 250 Hz within a bandwidth between 0.1 and 100 Hz. As linear features, absolute band power was obtained per clinical frequency band: delta (0.1~4 Hz), theta (4~8 Hz), alpha (8~12 Hz), beta (12~30 Hz) and gamma (30~100 Hz); considering five regions: prefrontal, frontal, central, parietal and occipital. As non-linear features, approximate entropy was calculated per channel and per clinical frequency band with addition of the broadband (0.1~100 Hz). Resulting feature vectors were grouped in line with the BIP outcome. Three groups were considered: low, moderate, and high pain. Finally, BIP-EEG patterns were classified in those three classes, achieving 96% accuracy. This result improves a previous work of a SVM classifier that used exclusively linear EEG features and showed an accuracy between 87% and 90% per class to predict central NP after spinal cord injury.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Data mining and big data methods - Inter-subject variability and personalized approaches, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Diagnosis and stratification of chronic pain patients is difficult due to a lack of sensitive biomarkers for altered nociceptive and pain processing. Recent developments enabled to preferentially stimulate epidermal nerve fibers and simultaneously quantify the psychophysical detection probability and neurophysiological EEG responses. In this work, we study whether using one or a combination of both outcome measures could aid in the observation of altered nociceptive processing in chronic pain. A set of features was extracted from data from a total of 66 measurements on 16 failed back surgery syndrome patients and 17 healthy controls. We assessed how well each feature discriminates both groups. Subsequently, we used a random forest classifier to study whether psychophysical features, EEG features or a combination can improve the classification accuracy. It was found that a classification accuracy of 0.77 can be achieved with psychophysical features, while a classification accuracy of 0.63 was achieved using only EEG features.

Keywords: Signal pattern classification
Abstract: In applications using electromyography (EMG), it is important to ensure high performance for all users (versatility among users) and to enable use without prior preparation (usability). Some of the current applications that use EMG normalize the signal through methods based on the measured maximum absolute value of EMG (maEMG), such as dynamic contraction (DC). However, usability is low when using DC because the reference value must be measured first every time the application is used. Further, the versatility among users is low because of the nonlinearity of EMG and the fact that maEMG varies among users. This study aimed to improve usability and versatility among users for continuous classification tasks using EMG. To this end, we developed a normalization method using sliding-window and z-score normalization techniques. The results reveal that the proposed method exhibits higher usability and versatility among users than DC. The proposed method does not require any calibration time, suggesting improved usability, while yielding the same classification accuracy as DC (57% for three target tasks) for a model trained using a subject’s own data. Further, for a model trained with other users’ data, the proposed method yields a classification accuracy of 53%, which is 18% higher than that of DC (35%), suggesting versatility among users. These results demonstrate that the proposed normalization method improves usability and versatility for users of practical applications that use EMG and perform continuous classification, such as prosthetic hands.

Keywords: Multivariate methods, Physiological systems modeling - Multivariate signal processing, Coupling and synchronization - Coherence in biomedical signal processing
Abstract: Different information-theoretic measures are available in the literature for the study of pairwise and higher-order interactions in multivariate dynamical systems. While these measures operate in the time domain, several physiological and non-physiological systems exhibit a rich oscillatory content that is typically analyzed in the frequency domain through spectral and cross-spectral approaches. For Gaussian systems, the relation between information and spectral measures has been established considering coupling and causality measures, but not for higher-order interactions. To fill this gap, in this work we introduce an information-theoretic framework in the frequency domain to quantify the information shared between a target process and two sources, even multivariate, and to highlight the presence of redundancy and synergy in the analyzed dynamical system. Firstly, we simulate different linear interacting processes by showing the capability of the proposed framework to retrieve amounts of information shared by the processes in specific frequency bands which are not detectable by the related time-domain measures. Then, the framework is applied on EEG time series representative of the brain activity during a motor execution task in a group of healthy subjects.

Keywords: Signal pattern classification, Adaptive filtering
Abstract: This paper aims to present an innovative approach based on Reinforcement Learning (RL) concept to detect contaminants’ type and minimize their effect on surface electromyography signal (sEMG). An agent-environment model was created based on the following elements: environment (muscle electrical activity), state (set of six features extracted from the signal), actions (application of filters/procedures to reduce the impact of each interference), and agent (controller, which will identify the type of contamination and take the appropriate action). The learning was conducted with Actor-Critic method. An average accuracy of 92.96% was achieved in an off-line experiment when detecting four contaminant types (electrocardiography (ECG) interference, movement artifact, power line interference, and additive white Gaussian noise).

Keywords: Signal pattern classification, Data mining and big data methods - Biosignal classification
Abstract: This study aims to classify rest and upper limb movements execution and intention using electroencephalogram (EEG) signals by developing machine-learning (ML) algorithms. Five different MLs are implemented, including k-Nearest Neighbor (KNN), Linear Discriminant Analysis (LDA), Naïve Bayes (NB), Support Vector Machine (SVM), and Random Forest (RF). The EEG data from fifteen healthy subjects during motor execution (ME) and motor imagination (MI) are pre-processed with Independent Component Analysis (ICA) to reduce eye-blinking associated artifacts. A sliding window technique varying from 1 s to 2 s is used to segment the signals. The majority voting (MV) strategy is employed during the post-processing stage. The results show that the application of ICA increases the accuracy of MI up to 6%, which is improved further by 1-2% using the MV (p<0.05). However, the improvement in the accuracies is more significant in MI (>5%) than in ME (<1%), indicating a more significant influence of eye-blinking artifacts in the EEG signals during MI than ME. Among the MLs, both RF and SVM consistently produced better accuracies in both ME and MI. Using RF, the 2 s window size produced the highest accuracies in both ME and MI than the smaller window sizes.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Data mining and big data methods - Biosignal classification, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Studies on intracranial electroencephalography (icEEG) recordings of patients with drug resistant epilepsy (DRE) show that epilepsy biomarkers propagate in time across brain areas. Here, we propose a novel method that estimates critical features of these propagations for different epilepsy biomarkers (spikes, ripples, and fast ripples), and assess their common onset as a reliable biomarker of the epileptogenic zone (EZ). For each biomarker, an automatic algorithm ranked the icEEG electrodes according to their timing occurrence in propagations and finally dichotomized them as onset or spread. We validated our algorithm on icEEG recordings of 8 children with DRE having a good surgical outcome (Engel score = 1). We estimated the overlap of the onset, spread, and entire zone of propagation with the resection (RZ) and the seizure onset zone (SOZ). Spike and ripple propagations were seen in all patients, whereas fast ripple propagations were seen in 6 patients. Spike, ripple, and fast ripple propagations had a mean duration of 28.3 ± 24.3 ms, 38.7 ± 37 ms, and 25 ± 14 ms respectively. Onset electrodes predicted the RZ and SOZ with higher specificity compared to the entire zone for all three biomarkers (p<0.05). Overlap of spike and ripple onsets presented a higher specificity than each separate biomarker onset: for the SOZ, the onsets overlap was more specific (97.89 ± 2.36) than the ripple onset (p=0.031); for the RZ, the onsets overlap was more specific (98.48 ± 1.5) than the spike onset (p=0.016). Yet, the entire zone for spike and ripple propagations predicted the RZ with higher sensitivity compared to each biomarker’s onset (or spread) (p<0.05). We present, for the first time, preliminary evidence from icEEG data that fast ripples propagate in time across large areas of the brain. The onsets overlap of spike and ripple propagations constitutes an extremely specific (but not sensitive) biomarker of the EZ, compared to areas of spread (and entire areas) in propagation.

Keywords: Signal pattern classification, Principal component analysis, Nonlinear dynamic analysis - Biomedical signals
Abstract: The recent development of novel multi-electrode recording technologies has revealed the existence of traveling patterns of cortical activity in many species and under different states of awareness. Among these, slow activation waves occurring under sleep and anesthesia have been widely investigated as they provide unique insights into network features such as excitability, connectivity, structure, and dynamics of the cerebral cortex. Such characterization is usually based on clustering methods which are constrained by a priori assumptions as to the number of clusters to be used or rely on wave-by-wave pattern reconstruction. Here, we introduce a new computational tool based on modal analysis of fluid flows which is robustly applied to multivariate electrophysiological data from cortical networks, namely the Energy-based Hierarchical Waves Clustering method (EHWC). EHWC is composed of three main steps: (1) detecting the occurrence of global waves; (2) reducing the data dimensionality via singular value decomposition; (3) clustering hierarchically the singled-out waves. The analysis does not require the single-channel contribution to the waves, which is a typical bottleneck in this kind of analysis due to the unavoidable intrinsic variability of locally recorded activity. For testing and validation, here we used in vivo extracellular recordings from mice cortex under three different levels of anesthesia. As a result, we found slow waves with an increasing number of propagation modes as the anesthesia level decreases, giving an estimate of the increasing complexity of network dynamics. This and other wave’s features replicate and extend the findings from previous literature, paving the way to extend the same approach to non-invasive electrophysiological recordings like EEG and fMRI used clinically for the characterization of brain dynamics and clinical stratification in brain lesions.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: Capturing the error perception of a human interacting with a Brain-Computer Interface (BCI) is a key piece in improving the accuracy of these systems and making the interaction more seamless. Convolutional Neural Networks (CNN) have recently been applied for this task rendering the model free of feature selection. We propose a new model with shorter temporal input trying to approximate its usability to that of a real-time BCI application. We evaluate and compare our model with some other recent CNN models using the Monitoring Error-Related Potential dataset, obtaining an accuracy of 80% with a sensitivity and specificity of 76% and 85%, respectively. These results outperform previous models. All models are made available online for reproduction and peer review.

Keywords: Principal component analysis
Abstract: Combining electroencephalography (EEG) to functional near-infrared spectroscopy (fNIRS) is a promising technique that has gained momentum thanks to their complementarity. While EEG measures the electrical activity of the brain, fNIRS records the variations in cerebral blood flow and related hemoglobin concentrations. However, both modalities are typically contaminated with artefacts. Muscle and eye artefacts, affect the EEG signals, while hemodynamic and oxygenation changes in the extracerebral compartment due to systemic changes (superficial layer) corrupt the fNIRS signals. Moreover, both signals are sensitive to sensor motion artefacts characterized by large amplitude. There are several well-established methods for removing artefacts for both modalities. The objective of this paper is to apply a common approach to denoise both EEG and fNIRS signals. Indeed Artifact Subspace Reconstruction (ASR) method, which is an automatic, online-capable and efficient method for deleting transient or large-amplitude EEG artefacts, can be a good alternative to also denoise fNIRS signals. In this paper, we first propose, a new more comprehensive formulation of ASR. Then, we study the effectiveness of the method in denoising both the EEG and fNIRS signals.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Electroencephalogram (EEG) signals have shown to be a good source of information for emotion recognition algorithms in Human-Brain interaction applications. In this paper, a reproducible framework is proposed for classifying human emotions based on EEG signals. The framework consists of extracting frequency-dependent features from raw EEG signals to form a three-dimensional EEG image which is classified by a convolutional neural network (CNN). The framework is used to show that the 3D input CNN outperforms conventional methods with two-dimensional input, using a public dataset. The implementation of the framework is publicly available to facilitate further work on this topic: https://github.com/KvanNoord/3D-CNN-EEG-Emotion-Classificat ion.

Keywords: Signal pattern classification, Data mining and big data methods - Biosignal classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Sleep staging is of paramount importance in children with suspicion of pediatric obstructive sleep apnea (OSA). Complexity, cost, and intrusiveness of overnight polysomnography (PSG), the gold standard, have led to the search for alternative tests. In this sense, the photoplethysmography signal (PPG) carries useful information about the autonomous nervous activity associated to sleep stages and can be easily acquired in pediatric sleep apnea home tests with a pulse oximeter. In this study, we use the PPG signal along with convolutional neural networks (CNN), a deep-learning technique, for the automatic identification of the three main levels of sleep: wake (W), rapid eye movement (REM), and non-REM sleep. A database of 366 PPG recordings from pediatric OSA patients is involved in the study. A CNN architecture was trained using 30-s epochs from the PPG signal for three-stage sleep classification. This model showed a promising diagnostic performance in an independent test set, with 78.2% accuracy and 0.57 Cohen’s kappa for W/NREM/REM classification. Furthermore, the percentage of time in wake stage obtained for each subject showed no statistically significant differences with the manually scored from PSG. These results were superior to the only state-of-the-art study focused on the analysis of the PPG signal in the automated detection of sleep stages in children suffering from OSA. This suggests that CNN can be used along with PPG recordings for sleep stages scoring in pediatric home sleep apnea tests.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Coupling and synchronization - Nonlinear coupling, Data mining and big data methods - Biosignal classification
Abstract: Epilepsy is frequently characterized by convulsive seizures, which are often followed by a postictal EEG suppression state (PGES). The ability to automatically detect and monitor seizure progression and postictal state can allow for early warning of seizure onset, timely intervention in seizures themselves, as well as identification of major complications in epilepsy such as status epilepticus and sudden unexpected death in epilepsy (SUDEP). To test whether it is possible to reliably differentiate these ictal and postictal states, we investigated 52 seizure records (both intracranial and scalp EEG) from 19 patients. Phase-amplitude cross-frequency coupling was calculated for each recording and used as an input to a convolutional neural network model, achieving the mean accuracy of 0.890.09 across all classes, with the worst class accuracy of 0.73 for one of the later ictal sub-states. When the trained model was applied to SUDEP patient data, it classified seizure recordings as primarily interictal and PGES-like state (70% and 26%, respectively), highlighting the fact that in SUDEP patients seizures primarily exist in postictal states and don’t show the ictal sub-state evolution. These results suggest that using frequency coupling markers with a machine learning algorithm can reliably identify ictal and postictal sub-states, which can open up opportunities for novel monitoring and management approaches in epilepsy.

Keywords: Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis, Connectivity, Coupling and synchronization - Coherence in biomedical signal processing
Abstract: There have been many previous studies on brain electrical activity and attention function, but the research on observing the cognitive function of attention from frequency brain electrical indicators still remains insufficient. This study proposed an attentional network test (ANT) of Chinese version and used frequency analysis methods to observe the power spectrum activity and functional connectivity of delta (δ), theta (θ), alpha (α) bands of EEG signals to further understand their relationship with attention networks. The attentional network test was composed of alerting network, orienting network and execute conflict network, and these networks were compared with the resting state in different frequency bands. The results showed that α band activity was significantly suppressed in all three attentional states, and the power of θ band activity dramatically increased for the execute conflict network. The negative connection of α band in the long distance (frontal lobe to parietal lobe or occipital lobe) might be a sign of resting state network, and the positive connections between δ and θ band in similar areas could be an indicator of execute conflict network. This pilot study suggests that the frequency domain analysis of EEG signals could be a great tool to visualize the brain activities in response to different attentional networks.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Neural networks and support vector machines in biosignal processing and classification, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Heart Rate Variability is a significant indicator of the Autonomic Neural System's functioning, traditionally evaluated from electrocardiogram recordings. Photoplethysmography sensors, like electrocardiograph devices, track the heart's activity and have been widely popularized by their use in smart watches and fitness trackers. In this study we develop a deep learning based approach which is able to successfully estimate the patient's Root Mean Square of the Successive Differences, a common heart rate variability metric, from lower quality, less expensive photoplethysmography sensors under a wide range of conditions.

Keywords: Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis, Time-frequency and time-scale analysis - Time-frequency analysis, Physiological systems modeling - Signal processing in physiological systems
Abstract: With the development of the Internet, the number of people with symptoms of Internet gaming disorder (IGD) has increased. In the past, psychologists used retrospective questionnaires to diagnose IGD, but retrospective questionnaires are difficult to diagnose IGD symptoms in time due to the requirement of more than 6 months of Internet gaming experience and the limitations of retrospective memory. Observing physiological regulation may help diagnose IGD in time. However, the observation of instantaneous physiological response is still limited by the lack of appropriate algorithms. Our previous study successfully combined complimentary ensemble empirical mode decomposition and normalized direct quadrature to obtain respiratory instantaneous frequency (IF) to overcome this limitation. This study further uses the game-related films as the stimulus materials to observe the difference in respiratory IF response per second of high-risk IGD (HIGD) and low-risk IGD (LIGD). The result showed that the respiratory IF of HIGD is lower than LIGD at the time of stimulation. In addition, the study also observes the dynamic change of respiratory IF per second (IFdiff). The results showed that the time point where there is a significant difference in IFdiff between HIGD and LIGD can be matched to the stimulation of the films. In summary, this study has showed that the IFdiff of HIGD and LIGD are different when stimulated and suggests that IFdiff may be used as a potential physiological marker to instantaneous distinguish and diagnosis the risk of IGD.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Physiological systems modeling - Signal processing in physiological systems, Signal pattern classification
Abstract: Continuous monitoring of patients with Parkinson’s Disease (PD) is critical for their effective management, as early detection of improvement or degradation signs play an important role on pharmaceutical and/or interventional plans. Within this work, a group of seven PD patients and a group of ten controls performed a set of exercises related to the evaluation of PD gait. Plantar pressure signals were collected and used to derive a set of analytics. Statistical tests and feature selection approaches revealed that the spatial distribution of the Center of Pressure during a static balance exercise is the most discriminative analytic and may be used for every-day monitoring of the patients. Results have revealed that out of the 28 features extracted from the collected signals, 10 were statistically significant (p < 0.05) and can be used to machine learning algorithms and/or similar approaches

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Coupling and synchronization - Coherence in biomedical signal processing
Abstract: The spectral method of cortico-muscular coherence (CMC) can reveal the communication patterns between the cerebral cortex and muscle periphery, thus providing guidelines for the development of new therapies for movement disorders and insights into fundamental motor neuroscience. The method is applied to electroencephalogram (EEG) and surface electromyogram (sEMG) recorded synchronously during a motor task. However, synchronous EEG and sEMG components are typically too weak compared to additive noise and background activities making significant coherence very difficult to detect. Dictionary learning and sparse representation have been proved effective in enhancing CMC levels. In this paper, we explore the potential of a recently proposed dictionary learning algorithm in combination with an improved component selection algorithm for CMC enhancement. The effectiveness of the method was demonstrated using neurophysiological data where it achieved considerable improvements in CMC levels.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Time-frequency and time-scale analysis - Time-frequency analysis, Time-frequency and time-scale analysis - Wavelets
Abstract: We proposed a sleep EEG-based brain age prediction model which showed higher accuracy than previous models. Six-channel EEG data were acquired for 6 hours sleep. We then converted the EEG data into 2D scalograms, which were subsequently inputted to DenseNet used to predict brain age. We then evaluated the association between brain aging acceleration and sleep disorders such as insomnia and OSA. The correlation between chronological age and expected brain age through the proposed brain age prediction model was 80% and the mean absolute error was 5.4 years. The proposed model revealed brain age increases in relation to the severity of sleep disorders. In this study, we demonstrate that the brain age estimated using the proposed model can be a biomarker that reflects changes in sleep and brain health due to various sleep disorders.

Keywords: Data mining and big data methods - Patient outcome and risk analysis, Neural networks and support vector machines in biosignal processing and classification, Physiological systems modeling - Signal processing in physiological systems
Abstract: Parkinson’s disease (PD) is a common neurodegenerative disease presenting with both motor and non-motor symptoms. Among PD motor symptoms, gait impairments are common and evolve over time. PD motor symptoms severity can be evaluated using clinical scales such as the Movement Disorder Society Unified Parkinson’s Rating Scale part III (MDS-UPDRS-III), which depend on the patient’s status at the time of assessment and are limited by subjectivity. Objective quantification of motor symptoms (i.e. gait) with wearable technology paired with Deep Learning (DL) techniques could help assess motor severity. The aims of this study were to: (i) apply DL techniques to wearable-based gait data to estimate MDS-UPDRS-III scores; (ii) test the DL approach on longitudinal dataset to predict the progression of MDS-UPDRS-III scores. PD gait was measured in the laboratory, during a 2 minute continuous walk, with a sensor positioned on the lower back. A DL Convolutional Neural Network (CNN) was trained on 70 PD subjects (mean disease duration: 3.5 years), validated on 58 subjects (mean disease duration: 5 years) and tested on 46 subjects (mean disease duration: 6.5 years). Model performance was evaluated on longitudinal data by quantifying the association (Pearson correlation (r)), absolute agreement (Intraclass correlation (ICC)) and mean absolute error between the predicted and true MDS-UPDRS-III. Results showed that MDS-UPDRS-III scores predicted with the proposed model, strongly correlated (r=0.82) and had a good agreement (ICC(2,1)=0.76) with true values; the mean absolute error for the predicted MDS-UPDRS-III scores was 6.29 points. The results from this study are encouraging and show that a DL-CNN model trained on baseline wearable-based gait data could be used to assess PD motor severity after 3 years.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: This paper presents an inception-based deep neural network for detecting lung diseases using respiratory sound input. Recordings of respiratory sound collected from patients are first transformed into spectrograms where both spectral and temporal information are well represented, in a process referred to as front-end feature extraction. These spectrograms are then fed into the proposed network, in a process referred to as back-end classification, for detecting whether patients suffer from lung-related diseases. Our experiments, conducted over the ICBHI benchmark metadataset of respiratory sound, achieve competitive ICBHI scores of 0.53/0.45 and 0.87/0.85 regarding respiratory anomaly and disease detection, respectively.

Keywords: Nonlinear dynamic analysis - Biomedical signals, Physiological systems modeling - Signal processing in physiological systems
Abstract: The Detrended Fluctuation Analysis (DFA) is a popular method for quantifying the self-similarity of the heart rate that may reveal complexity aspects in cardiovascular regulation. However, the self-similarity coefficients provided by DFA may be affected by an overestimation error associated with the shortest scales. Recently, the DFA has been extended to calculate the multifractal-multiscale self-similarity and some evidence suggests that overestimation errors may affect different multifractal orders. If this is the case, the error might alter substantially the multifractal-multiscale representation of the cardiovascular self-similarity. The aim of this work is 1) to describe how this error depends on the multifractal orders and scales and 2) to propose a way to mitigate this error applicable to real cardiovascular series.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Signal pattern classification, Time-frequency and time-scale analysis - Wavelets
Abstract: Abstract — Motility of the gastrointestinal tract (GI) is governed by an bioelectrical event termed slow waves. Accurately measuring the characteristics of GI slow waves is critical to understanding its role in clinical applications. High-resolution (HR) bioelectrical mapping involves placing a spatially dense array of electrodes directly over the surface of the GI wall to record the spatiotemporal changes in slow waves. A micro-electrode array (MEA) with spatial resolution of 200 μm in an 8x8 configuration was employed to record intestinal slow waves using isolated tissues from small animals including rodents, shrews and ferrets. A filtering, processing, and analytic pipeline was developed to extract useful metrics from the recordings. The pipeline relied on CWT and Hilbert Transform to identify the frequency and phase of the signals, from which the individual activation times of slow waves were identified and clustered using k-means. A structural similarity index was applied to group the major activation patterns. Overall, the pipeline identified 91 cycles of slow waves from 300 s of recordings in mice, with an average frequency of 20.68 ± 0.71 cpm, amplitude of 7.94 ± 2.15 µV, and velocity of 3.64 ± 1.75 mm s-1. Three major propagation patterns were identified during this period. The findings of this study will inform the development of a high throughput software platform for future in vitro pharmacological studies using the MEA. Clinical Relevance — The proposed analysis pipeline will be used to quantify response of GI tissues to pharmacological agents.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Nonlinear dynamic analysis - Biomedical signals
Abstract: For the last decades, ripples 80-200Hz (R)and fast ripples 200-500Hz (FR) were intensively studied as biomarkers of the epileptogenic zone (EZ). Recently, Very fast ripples 500-1000Hz (VFR) and ultra-fast ripples 1000-2000Hz (UFR) recorded using standard clinical macro electrodes have been shown to be more specific for EZ. High-sampled microelectrode recordings can bring new insights into this phenomenon of high frequency, multiunit activity. Unfortunately, visual detection of such events is extremely time consuming and unreliable. Here we present a detector of ultra-fast oscillations (UFO, >1kHz). In an example of two patients, we detected 951 UFOs which were more frequent in epileptic (8.6/min) vs. non-epileptic hippocampus (1.3/min). Our detection method can serve as a tool for exploring extremely high frequency events from microelectrode recordings.

Keywords: Data mining and big data methods - Machine learning and deep learning methods
Abstract: Ballistocardiagram (BCG) is a non-contact and non-invasive technique to obtain physiological information with the potential to monitor Cardio Vascular Disease (CVD) at home. Accurate detection of J-peak is the key to get critical indicators from BCG signals. With the development of deep learning methods, many researches have applied convolution neural network (CNN) and recurrent neural network (RNN) based models in J-peak detection. However, these deep learning methods have limitations in inference speed and model complexity. To improve the computational efficiency and memory utilization, we propose a robust lightweight neural network model, called JwaveNet. Moreover, in the preprocessing stage, J-peaks are re-modeled by a new transformation method based on their physiological meaning, which has been proven to increase performance. In our experiment, BCG signals, including four different sleeping positions, were collected from 24 subjects with synchronous electrocardiogram (ECG) signals. The experiment results have shown that our lightweight model greatly reduces latency and model size compared to other baseline models with high detecting accuracy.

Keywords: Signal pattern classification, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Electroencephalogram (EEG) is a crucial tool in the diagnosis and management of epilepsy. The process of analyzing EEG is time consuming leading to the development of seizure detection algorithms to aid its analysis. This approach is limited since it requires seizures to occur during monitoring periods and can often lead to misdiagnosis in cases where seizure occurrence is rare. For such cases, it has been shown that the interictal periods in EEG signals, which is the predominant state in long-term monitoring, can be useful for the diagnosis of epilepsy. This paper presents an algorithm, using the information in interictal periods, to discriminate between long-term EEG recordings of epilepsy patients and healthy subjects. It extracts several time and frequency-time domain features from the signals and classifies them using an ensemble classifier, achieving 100% sensitivity and 98.7% specificity in classifying 267 recordings from 105 subjects. The results demonstrate the feasibility of this approach to reliably identify EEG recordings of epilepsy subjects automatically which can be highly useful to facilitate screening and diagnosis of epilepsy, especially in those parts of the world where there is a lack of trained personnel for interpreting EEG signals.

Keywords: Multivariate methods
Abstract: This study explores the feasibility of implementation of an analysis framework of neonatal EEG, including ML, sonification and intuitive visualization, on a low power IoT edge device. Electroencephalography (EEG) analysis is a very important tool to detect brain disorders. Neonatal seizure detection is a known, challenging problem. Under-resourced communities across the globe are particularly affected by the cost associated with EEG analysis and interpretation. Machine learning (ML) techniques have been successfully utilized to automate seizure detection in neonatal EEG, in order to assist a healthcare professional in visual analysis. Several usage scenarios are reviewed in this study. It is shown that both sonification and ML can be efficiently implemented on low-power edge platforms without any loss of accuracy. The developed platform can be easily expanded to address EEG analysis applications in neonatal and adult population.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Pattern recognition, Data mining and big data methods - Patient outcome and risk analysis
Abstract: Computer-vision techniques provide a way to conduct low-cost, portable, and real-time evaluations of exercises performed as a part of physical rehabilitation. Recent data-driven methods have explored using deep learning on 3D body-landmark sequences for automatic assessment of physical rehabilitation exercises. However, existing deep learning methods using convolutional neural networks (CNN) fail to utilize the spatial connection information of the human body, which limits the accuracy of these assessments. To overcome these limitations and provide a more accurate method to assess physical rehabilitation exercises, we propose a deep learning framework using a graph convolutional network (GCN) with self-supervised regularization. The experimental results on an existing benchmark dataset validate that the proposed method achieves state-of-the-art performance with lower error than other CNN methods, and the self-supervised learning improves the prediction accuracy.

Keywords: Nonlinear dynamic analysis - Biomedical signals, Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: Cardiac auscultation is the key screening procedure to detect and identify cardiovascular diseases (CVDs). One of many steps to automatically detect CVDs using auscultation, concerns the detection and delimitation of the heart sound boundaries, a process known as segmentation. Whether to include or not a segmentation step in the signal classification pipeline is nowadays a topic of discussion. Up to our knowledge, the outcome of a segmentation algorithm has been used almost exclusively to align the different signal segments according to the heartbeat. In this paper, the need for a heartbeat alignment step is tested and evaluated over different machine learning algorithms, including deep learning solutions. From the different classifiers tested, Gate Recurrent Unit (GRU) Network and Convolutional Neural Network (CNN) algorithms are shown to be the most robust. Namely, these algorithms can detect the presence of heart murmurs even without a heartbeat alignment step. Furthermore, Support Vector Machine (SVM) and Random Forest (RF) algorithms require an explicit segmentation step to effectively detect heart sounds and murmurs, the overall performance is expected drop approximately 5% on both cases.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Nonlinear dynamic analysis - Biomedical signals
Abstract: Extensive efforts have been recently devoted to implement fast and reliable algorithms capable of assessing the physiological response of the organism to physiological stress. In this study, we propose the comparison between model-free and linear parametric methods as regards their ability to detect alterations in the dynamics and in the complexity of cardiovascular and respiratory variability evoked by postural and mental stress. Dynamic entropy (DE) and information storage (IS) measures were calculated on three physiological time-series, i.e. heart period, respiratory volume and systolic arterial pressure, on 61 healthy subjects monitored in resting conditions as well as during head-up tilt and while performing a mental arithmetic task. The results of the comparison suggest the feasibility of DE and IS measures computed from different physiological signals to discriminate among resting and stress states. If compared to the model-free algorithm, the faster linear method appears to be capable of detecting the same (or even more) statistically significant variations of DE or IS between resting and stress conditions, being thus in perspective more suitable for the integration within wearable devices. The computation of entropy indices extracted from multiple physiological signals acquired through wearables will allow a real-time stress assessment on people during daily-life situations.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Abstract— To date, spinal problems are not rare, and relevant therapies are always required. Although the combination of photoacoustic imaging (PAI) and spinal fusion surgery, a widely applied operation for spinal cures, is unprecedented, we assume that such combination might improve the accuracy and safety of the surgery. This paper aims to testify that PAI is effective in monitoring and navigating during the spinal fusion surgery operation. Specifically, we examined the optical absorption spectrum of bones to determine the optimal laser wavelength as 532nm. Afterwards, we measured the photoacoustic signals of this bone samples, discovering that the signals of two kinds of samples, cortical bone and cancellous bone, differ considerably in frequency domain. It demonstrated the feasibility that PAI is effective enough to distinguish different bone tissues during the spinal fusion surgery.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Measuring the respiration and heart rate unobtrusively in home settings is an important goal for health monitoring. In this work, use of a pressure sensitive mat was explored. Two methods using body morphology information, based on shoulder blades and weighted centroid, were developed for respiration rate (RR) calculation. Heart rate (HR) was calculated by combining the frequency information from different body regions. Experimental data were collected from 15 participants in a supine position via a pressure sensitive mat placed under the upper torso. RR and HR estimations derived from accelerometer sensors attached to participants’ bodies were used as references to evaluate the accuracy of the proposed methods. All three methods achieved a reasonable estimation compared to the reference. The root mean squared error of the proposed RR estimation methods were 1.32 and 0.87 breath/minute respectively, and the root mean squared error of the HR estimation method was 5.55 bpm.

Keywords: Signal pattern classification, Data mining and big data methods - Machine learning and deep learning methods, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Traumatic brain injury (TBI) is a sudden injury that causes damage to the brain. TBI can have wide-ranging physical, psychological, and cognitive effects. TBI outcomes include acute injuries, such as contusion or hematoma, as well as chronic sequelae that emerge days to years later, including cognitive decline and seizures. Some TBI patients develop post-traumatic epilepsy (PTE), or recurrent and unprovoked seizures following TBI. In recent years, significant efforts have been made to identify biomarkers of epileptogenesis, the process by which a normal brain becomes capable of generating seizures. These biomarkers would allow for a higher standard of care by identifying patients at risk of developing PTE as candidates for antiepileptogenic interventions. In this paper, we use deep neural network architectures to automatically detect potential biomarkers of PTE from electroencephalogram (EEG) data collected between post-injury day 1-7 from patients with moderate-to-severe TBI. Continuous EEG is often part of multimodal monitoring for TBI patients in intensive care units. Clinicians review EEG to identify the presence of epileptiform abnormalities (EAs), such as seizures, periodic discharges, and abnormal rhythmic delta activity, which are potential biomarkers of epileptogenesis. We show that a recurrent neural network trained with continuous EEG data can be used to identify EAs with the highest accuracy of 80.78%, paving the way for robust, automated detection of epileptiform activity in TBI patients.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: Attention can be measured by different types of cognitive tasks, such as Stroop, Eriksen Flanker, and Psychomotor Vigilance Task (PVT). Despite the differing content of the three cognitive tasks, they all require the use of visual attention. To learn the generalized representations from the electroencephalogram (EEG) of different cognitive attention tasks, extensive intra and inter-task attention classification experiments were conducted on three types of attention task data using SVM, EEGNet, and TSception.With cross-validation in intra-task experiments, TSception has significantly higher classification accuracies than other methods, achieving 82.48%, 88.22%, and 87.31% for Stroop, Flanker, and PVT tests respectively. For inter-task experiments, deep learning methods showed superior performance over SVM with most of the accuracy drops not being significant. Our experiments indicate that there is common knowledge that exists across cognitive attention tasks, and deep learning methods can learn generalized representations better.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Heart rate (HR) and respiratory rate (RR) are very important physiological variables useful to evaluate the cardiorespiratory system. At present, there is a great interest by the general population in knowing their health status, quickly and easily. Accordingly, several approaches have been proposed to achieve that goal. In this study, the simultaneous estimation of the instantaneous HR and RR values was achieved by the image photoplethysmography (iPPG) technique, in the contact mode directly implemented in a smartphone. iPPG results were compared with those obtained using specialized biomedical sensors such as the electrocardiogram and the respiratory effort band. Performance evaluation included three different respiratory maneuvers in five healthy volunteers. The absolute mean error for instantaneous HR and RR estimations reached 0.94 ± 0.28 beats per minute and 0.40 ± 0.11 breaths per minute, respectively. The mean correlation index was 0.69 ± 0.14 between the iPPG-derived respiratory signal and the respiratory effort reference signal. These results appear to indicate that the contact iPPG method implemented directly on the smartphone is a good option, accessible to the common population to estimate the instantaneous HR and RR values outside specialized clinical environments, e.g., in the point-of-contact office.

Keywords: Connectivity, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: The main objective of this study was to examine the influence that recording length, sampling frequency, and imaging modality have on the estimation and characterization of spontaneous brain meta-states during rest. To this end, a recently developed method of meta-state extraction and characterization was applied to a subset of 16 healthy elderly subjects from two independent electroencephalographic and magnetoencephalographic (EEG/MEG) databases. The recordings were segmented into the first 5, 10, 15, 20, 25, 30, 60 and 90-s of artifact-free activity and meta-states were extracted. Temporal activation sequence (TAS) complexity, which characterizes the complexity of the metastateactivation sequences during rest, was calculated. Then, its stability as a function of segment length, sampling frequency, and imaging modality was assessed. The results showed that, in general, the minimum segment length needed to fully characterize resting-state meta-state activation complexity ranged from 15 to 25 seconds. Moreover, it was found that the sampling frequency has a slight influence on the complexity measure, and that results were similar across EEG and MEG. The findings indicate that the proposed methodology can be applied to both EEG and MEG recordings and displays stable behavior with relatively short segments. However, methodological choices, such as sampling frequency, should be carefully considered.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification
Abstract: Deep learning methods, and in particular Convolutional Neural Networks (CNNs), have shown breakthrough performance in a wide variety of classification applications, including electroencephalogram-based Brain Computer Interfaces (BCIs). Despite the advances in the field, BCIs are still far from the subject-independent decoding of brain activities, primarily due to substantial inter-subject variability. In this study, we examine the potential application of an ensemble CNN classifier to integrate the capabilities of CNN architectures and ensemble learning for decoding EEG signals collected in motor imagery experiments. The results prove the superiority of the proposed ensemble CNN in comparison with the average base CNN classifiers, with an improvement up to 9% in classification accuracy depending on the test subject. The results also show improvement with respect to the performance of a number of state-of-the-art methods that have been previously used for subject-independent classification in the same datasets used here (i.e., BCI Competition IV 2A and 2B datasets).

Keywords: Independent component analysis
Abstract: This paper proposes higher-order tensor independent component analysis (HOT-ICA). HOT-ICA is a tensor ICA that makes effective use of the relationships among the axes of a separating tensor. We deal with multiple-target signal separation in a multiple-input multiple-output (MIMO) radar system to detect respiration and heartbeat. Numerical physical experiments demonstrate the significance of the HOT-ICA which keeps the tensor structure unchanged to fully utilizes the high-dimensionality of the separation tensor.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Principal component analysis, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Seizure detection and seizure-type classification are best performed using intra-cranial or full-scalp electroencephalogram (EEG). In embedded wearable systems however, recordings from only a few electrodes are available, reducing the spatial resolution of the signals to a handful of timeseries at most. Taking this constraint into account, we tested the performance of multiple classifiers using a subset of the EEG recordings by selecting a single trace from the montage or performing a dimensionality reduction over each hemispherical space. Our results support that Random Forest (RF) classifiers lead most efficient and stable classification performances over Support Vector Machines (SVM). Interestingly, tracking the feature importances using permutation tests reveals that classical EEG spectrum power bands display different rankings across the classifiers: low frequencies (delta, theta) are most important for SVMs while higher frequencies (alpha, gamma) are more relevant for RF and Decision Trees. We reach up to 94.3% +/- 5.3% accuracy in classifying absence from tonic-clonic seizures using state-of-art sampling methods for unbalanced datasets and leave-patients-out 3-fold cross-validation policy.

Keywords: Time-frequency and time-scale analysis - Nonstationary analysis and modeling, Principal component analysis, Signal pattern classification
Abstract: Despite the recent advancements in the development of compact, low-density wearable EEG devices, EEG artifacts, especially motion artifacts severely affect the performance of such devices in real-time environments. A promising solution is Artifacts Subspace Reconstruction (ASR), an online capable, component-based approach compatible with real-time computations that can automatically remove non-stationary, large-amplitude artifacts in EEG data. However, ASR has been validated only on high-density EEG datasets and it is unclear whether it is equally efficient on data recorded with portable low-density EEG devices, as well.

Here, we validated ASR from SSVEPs elicited by sinusoidal checkerboard pattern stimulation at multiple frequencies (2, 4, and 8 Hz) recorded from six adult subjects wearing a wireless EEG system based on BioWolf, an 8-channel ultra-low-power brain-computer interfaces (BCI) platform. Specifically, we recorded the data in two experimental conditions namely emph{Artifact-Free} and emph{Artifact} to formally evaluate the performance of ASR. In addition, we systematically varied ASR rejection threshold parameter (k) and processing mode (correction or removal of artifacted segments) in an attempt to characterize these crucial parameters.

Empirical results show that 1) Even with such a few channels, ASR efficiently corrects artifacts, with an overall improvement of 34.23% in SSVEP quality in emph{Artifact} trials compared to SSVEPs computed without ASR processing; 2) While ASR artifact correction performs better for low frequency (2 Hz) with an improvement of 18.68%, ASR artifact removal is more efficient for higher frequencies (4 and 8 Hz) improving SSVEPs by 67.5% and 49.5% respectively; 3) Optimal (k) values are found to be in the range established for high-density montages; 4) By manually choosing the optimal ASR parameters on single subjects we found an improvement in SSVEP quality as high as 45.23%. We conclude that ASR is suitable for

Keywords: Signal pattern classification, Physiological systems modeling - Signal processing in simulation
Abstract: Canonical correlation analysis (CCA) is one of the most used algorithms in the steady-state visual evoked potentials (SSVEP)-based brain-computer interface (BCI) systems due to its simplicity, efficiency, and robustness. Researchers have proposed modifications to CCA to improve its speed, allowing high-speed spelling and thus a more natural communication. In this work, we combine two approaches, the filter-bank (FB) approach to extract more information from the harmonics, and a range of different supervised methods which optimize the reference signals to improve the SSVEP detection. The proposed models are tested on the publicly available benchmark dataset for SSVEP-based BCIs and the results show improved performance compared to the state-of-the-art methods and, in particular, the proposed FBMwayCCA approach achieves the best results with an information transfer rate (ITR) of 134.8±8.4 bits/minute. This study indeed suggests the feasibility of combining the fundamental and harmonic SSVEP components with supervised methods in target identification to develop high-speed BCI spellers.

Keywords: Directionality, Physiological systems modeling - Signal processing in physiological systems, Physiological systems modeling - Signal processing in simulation
Abstract: We present the implementation to cardiovascular variability of a method for the information-theoretic estimation of the directed interactions between event-based data. The method allows to compute the transfer entropy rate (TER) from a source to a target point process in continuous time, thus overcoming the severe limitations associated with time discretization of event-based processes. In this work, the method is evaluated on coupled cardiovascular point processes representing the heartbeat dynamics and the related peripheral pulsation, first using a physiologically-based simulation model and then studying real point-process data from healthy subjects monitored at rest and during postural stress. Our results document the ability of TER to detect direction and strength of the interactions between cardiovascular processes, also highlighting physiologically plausible interaction mechanisms.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Time-frequency and time-scale analysis - Time-frequency analysis, Physiological systems modeling - Signal processing in physiological systems
Abstract: Respiratory diseases are among the leading causes of death worldwide. Preventive measures are essential to avoid and increase the odds of a successful recovery. An important screening tool is pulmonary auscultation, an inexpensive, noninvasive and safe method to assess the mechanics and dynamics of the lungs. On the other hand, it is a difficult task for a human listener since some lung sound events have a spectrum of frequencies outside of the human hearing ability. Thus, computer assisted decision systems might play an important role in the detection of abnormal sounds, such as crackle or wheeze sounds. In this paper, we propose a novel system, which is not only able to detect abnormal lung sound events, but it is also able to classify them. Furthermore, our system was trained and tested using the publicly available ICBHI 2017 challenge dataset, and using the metrics proposed by the challenge, thus making our framework and results easily comparable. Using a Mel Spectrogram as an input feature for our convolutional neural network, our system achieved results in line with the current state of the art, an accuracy of 43%, and a sensitivity of 51%.

Keywords: Data mining and big data methods - Pattern recognition, Data mining and big data methods - Biosignal classification, Data mining and big data methods - Inter-subject variability and personalized approaches
Abstract: Patients suffering from pulmonary diseases typically exhibit pathological lung ventilation in terms of homogeneity. Electrical Impedance Tomography (EIT) is a non-invasive imaging method that allows to analyze and quantify the distribution of ventilation in the lungs. In this article, we present a new approach to promote the use of EIT data and the implementation of new clinical applications for differential diagnosis, with the development of several machine learning models to discriminate between EIT data from healthy and non-healthy subjects. EIT data from 16 subjects were acquired: 5 healthy and 11 non-healthy subjects (with multiple pulmonary conditions). Preliminary results have shown accuracy percentages of 66% in challenging evaluation scenarios. The results suggest that the pairing of EIT feature engineering methods with machine learning methods could be further explored and applied in the diagnostic and monitoring of patients suffering from lung diseases. Also, we introduce the use of a new feature in the context of EIT data analysis (Impedance Curve Correlation).

Keywords: Adaptive filtering, Data mining and big data methods - Biosignal classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: The purpose of computer-aided diagnosis (CAD) systems is to improve the detection of diseases in a shorter time and with reduced subjectivity. A robust system frequently requires a noise-free input signal. For CADs which use heart sounds, this problem is critical as heart sounds are often low amplitude and affected by some unavoidable sources of noise such as movement artefacts and physiological sounds. Removing noises by using denoising algorithms can be beneficial in improving the diagnostics accuracy of CADs. In this study, four denoising algorithms were investigated. Each algorithm has been carefully adapted to fit the requirements of the phonocardiograph signal. The effect of the denoising algorithms was objectively compared based on the improvement it introduces in the classification performance of the heart sound dataset. According to the findings, using denoising methods directly before classification decreased the algorithm's classification performance because a murmur was also treated as noise and suppressed by the denoising process. However, when denoising using Wiener estimation based spectral subtraction was used as a preprocessing step to improve the segmentation algorithm, it increased the system's classification performance with a sensitivity of 96.0%, a specificity of 74.0%, and an overall score of 85.0%. As a result, to improve performance, denoising can be added as a preprocessing step into heart sound classifiers that are based on heart sound segmentation.

Keywords: Signal pattern classification, Nonlinear dynamic analysis - Biomedical signals, Physiological systems modeling - Multivariate signal processing
Abstract: The electromyography (EMG) signals have been widely utilized in human–robot interaction for extracting user hand/arm motion instructions. A major challenge of the online interaction with robots is the reliable EMG recognition from real-time data. However, previous studies mainly focused on using steady-state EMG signals with a small number of grasp patterns to implement classification algorithms, which is insufficient to generate robust control regarding the dynamic muscular activity variation in practice. Introducing more EMG variability during training and validation could implement a better dynamic-motion detection, but only limited research focused on such grasp-movement identification, and all of those assessments on the non-static EMG classification require supervised ground-truth label of the movement status. In this study, we propose a framework for classifying EMG signals generated from continuous grasp movements with variations on dynamic arm/hand postures, using an unsupervised motion status segmentation method. We collected data from large gesture vocabularies with multiple dynamic motion phases to encode the transitions from one intent to another based on common sequences of the grasp movements. Two classifiers were constructed for identifying the motion-phase label and grasp-type label, where the dynamic motion phases were segmented and labeled in an unsupervised manner. The proposed framework was evaluated in real-time with the accuracy variation over time presented, which was shown to be efficient due to the high degree of freedom of the EMG data.

Keywords: Time-frequency and time-scale analysis - Nonstationary analysis and modeling, Physiological systems modeling - Signal processing in physiological systems, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Non-contact blood pressure (BP) estimation with imaging photoplethysmogram (PPG) that can be acquired by camera is a promising alternative to cuff-based technology because of its nature of pervasive, low-cost, and being continuous. Most of the non-contact BP estimation methods are based on the principle of pulse transit time (PTT) as being used for wearable cuffless BP measurement. However, PTT-based method on the one hand requires simultaneous capture of images of multiple skin sites with the sites being at a distance from each other; and on the other hand, it can only partially reflect BP changes according to previous studies. In this paper, we propose to use a different camera PPG feature that has not yet been fully studied – pulse width at half amplitude (PWHA) for the evaluation of BP in a non-contact way. PWHA can be obtained from a single-site camera PPG, and it can indicate BP changes. The relationship of PWHA and BP was analyzed on 16 healthy subjects with BP changes induced by deep breathing and stepping exercise. The results showed that beat-to-beat PWHA can well track dynamic BP changes, and it is inversely related to BP across the sampled population and within each individual with about 80% individuals having high correlations. The findings suggest that PWHA can reflect the dynamic changes in cardiovascular characteristics and thereby BP changes, demonstrating the feasibility of imaging PWHA for non-contact BP estimation beyond the PTT method.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods, Signal pattern classification
Abstract: Cardiovascular disease (CVD) is a serial of diseases with global leading causes of death. Electrocardiogram (ECG) is the most commonly used basis for CVD diagnosis due to its low cost and no injury. Due to the great performance shown in classification tasks with large-scale data sets, deep learning has been widely applied in ECG diagnosis. Manual labeling is a time-consuming and labor-intensive job, which makes it error-prone and easy to labeled wrongly. These noisy labels cause deterioration in performance since deep neural network is easy to over-fitting with noisy labels. However, currently, only limited studies have been concerned with this problem. To alleviate the performance degradation caused by noisy labels, we come up with an optimization method combining data clean and anti-noise loss function. Our method filters the noisy data by data-clean method, followed by training the network with boot-hard loss function. The experiment is carried on MIT-BIH arrhythmia database and we take a 1-D CNN model for test. The result indicates that our optimization method can produce an effective improvement for noisy label problems when the proportion of incorrect labels ranging from 10% to 50%.

Keywords: Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis, Adaptive filtering
Abstract: The use of ECG in cardiovascular health monitoring is well established. The signal is collected using specialised equipment, capturing the electrical discharge properties of the human heart. This produces a well-structured signal trace, which can be characterised through its peaks and troughs. The signal can then be used by clinicians to diagnose cardiac disorders. However, as with any measuring equipment, the ECG output signal can experience deterioration resulting from noise. This can happen due to environmental interference, human issues or measuring equipment failure, necessitating the development of various denoising strategies to reduce, or remove, the noise. In this paper, we study typically occurring types of noise and implement popular strategies used to rectify them. We also show, that the given strategy's denoising potential is directly related to R-wave detection, and provide best strategies to apply when faced with specific noise type.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Pattern recognition
Abstract: Eye dynamics, a typical expression of brain activities, is an emerging modality for emerging and promising smart health applications. Electrooculogram (EOG) – a natural bio-electric signal generated during eye movements, if decoded, is of great potential to reveal the user’s mind and enable voice-free communication for patients with amyotrophic lateral sclerosis (ALS). ALS patients usually lose physical movement abilities including speech and handwriting but fortunately can move their eyes. In this study, we propose a novel deep transfer learning-empowered system, called “eyeSay”, which leverages both deep learning and transfer learning for intelligent eye EOG-to-speech translation. More specifically, we have designed a multi-stage convolutional neural network (CNN) to analyze the eye-written words, named as CNN-word. Moreover, to reveal fundamental patterns of eye movements, we build a transferable feature extractor, CNN-stroke, upon eye strokes that are building components of an eye word. Then, we transfer the CNN-stroke model to the eye word learning task in an innovative way, that is, use CNN-stroke as an additional branch of CNN-word to generate a stroke probability map. The achieved boostCNN-word model, enhanced by the transferable feature extractor, has greatly improved the eye word decoding performance. This novel study will directly contribute to voice-free communications for ALS patients, and greatly advance the ubiquitous eye EOG-based smart health area.

Keywords: Nonlinear dynamic analysis - Biomedical signals, Signal pattern classification, Physiological systems modeling - Signal processing in physiological systems
Abstract: There is growing evidence that seizures are accompanied by multi-system changes, not only in the brain but also in organs and systems under its control. Non-EEG measurements from these systems could be leveraged to improve seizure prediction, which is difficult but critical to the success of next-generation epilepsy therapies. Clinical electrophysiology studies during presurgical patient evaluations routinely collect continuous EEG but also ECG data that span multiple days. Prior work has reported electrocardiographic changes but has primarily focused on ventricular activity and brief peri-ictal intervals. Using novel data-driven classification and separation of the ECG high-dimensional signal space, this study investigated seizure-related changes in both ventricular and atrial activity. Measures of complexity as well as heart rate and R-R interval length were analyzed over time in continuous ECGs from 22 pediatric patients with pharmacoresistant seizures and no diagnosed cardiovascular anomalies. Fifteen patients (>68%) had significant changes in atrial or ventricular activity (or both) in intervals containing seizures. Thus, for a substantial number of patients, cardiac markers may be specifically modulated by seizures and could be leveraged to improve and personalize seizure prediction.

Clinical Relevance— Electrocardiographic changes during seizure evolution in children with medically refractory epilepsy remain relatively unexplored. Using continuous single-lead ECG recordings (median = 93.3 h) from 22 pediatric patients with medically refractory epilepsy, seizure-related changes in atrial signal complexity and/or ventricular parameters (including heart rate and R-R interval length) were identified. These may represent novel non-EEG, markers of seizure evolution that could be used to ultimately improve next-generation targeted therapies.

Keywords: Data mining and big data methods - Patient outcome and risk analysis, Physiological systems modeling - Signal processing in physiological systems, Connectivity
Abstract: Autocorrelation in functional MRI (fMRI) time series has been studied for decades, mostly considered as noise in the time series which is removed via prewhitening with an autoregressive model. Recent results suggest that the coefficients of an autoregressive model fit to fMRI data may provide an indicator of underlying brain activity, suggesting that prewhitening could be removing important diagnostic information. This paper explores the explanatory value of these autoregressive features extracted from fMRI by considering the use of these features in a classification task. As a point of comparison, functional network based features are extracted from the same data and used in the same classification task. We find that in most cases, network based features provide better classification accuracy. However, using principal component analysis to combine network based features and autoregressive features for classification based on a support vector machine provides improved classification accuracy compared to single features or network features, suggesting that when properly combined there may be additional information to be gained from autoregressive features.

Keywords: Data mining and big data methods - Biosignal classification, Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Pattern recognition
Abstract: Sedentary behavior is considered as a major public health challenge, linked with many chronic diseases and premature mortality. In this paper, we propose a steps counting -based machine learning approach for the prediction of sedentary behavior. Our work focuses on analyzing historical data from multiple users of wearable physical activity trackers and exploring the performance of four machine learning algorithms, i.e., Logistic Regression, Random Forest, XGBoost, Convolutional Neural Networks, as well as a Majority Vote Ensemble of the algorithms. To train and test our models we employed a crowd sourced dataset containing a month’s data of 33 users. For further evaluation, we employed a dataset containing 6 months of data of an additional user. The results revealed that while all models succeed in predicting next-day sedentary behavior, the ensemble model outperforms all baselines, as it manages to predict sedentary behavior and reduce false positives more effectively. On the multi-subjects test dataset, our ensemble model achieved an accuracy of 82.12% with a sensitivity of 74.53% and a specificity of 85.71%. On the additional unseen dataset, we achieved 76.88% in accuracy, 63.27% in sensitivity and 81.75% in specificity. These outcomes provide the ground towards the development of real-life artificially intelligent systems for sedentary behavior prediction.

Keywords: Principal and independent component analysis - Blind source separation, Physiological systems modeling - Multivariate signal processing
Abstract: Unobtrusive mental state monitoring based on neurosphysiological signals has seen thriving developments over the past decade, with a wide area of applications, from rehabilitation to neuroergonomics and neuromarketing. Particularly, electroencephalography (EEG) and electrooculography (EOG) have been popular techniques to obtain cognitive-relevant biosignals. However, current wearable systems may still pose practical inconvenience, motivating further interest to integrate EOG+EEG recording into streamlined frontal-only sensor montages with sufficient signal fidelity. We propose, here, a spatial filtering approach to reliably extract EOG signals from a reduced set of frontal EEG electrodes, placed on non-hair-bearing (NHB) areas. Within a common signal analytic framework, two distinct schemes are examined. The one is based on standard linear least squares (LLS) and the other on Least Absolute Shrinkage and Selection Operator (LASSO). Both schemes are data-driven techniques, require a small amount of training data, and lead to reliable estimators of EOG activity from EEG signals. The LASSO-based technique, in addition, provides guidelines that generalize well across subjects. Using experimental data, we provide some empirical evidence that our estimators can replace the actual EOG signals in algorithmic pipelines that automatically detect oculographic events, like blinks and saccades.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Signal pattern classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Abstract— Artifact detection and removal is a crucial step in all data preprocessing pipelines for physiological time series data, especially when collected outside of controlled experimental settings. The fact that such artifact is often readily identifiable by eye suggests that unsupervised machine learning algorithms may be a promising option that do not require manually labeled training datasets. Existing methods are often heuristic-based, not generalizable, or developed for controlled experimental settings with less artifact. In this study, we test the ability of three such unsupervised learning algorithms, isolation forests, 1-class support vector machine, and K-nearest neighbor distance, to remove heavy cautery-related artifact from electrodermal activity (EDA) data collected while six subjects underwent surgery. We first defined 12 features for each half-second window as inputs to the unsupervised learning methods. For each subject, we compared the best performing unsupervised learning method to four other existing methods for EDA artifact removal. For all six subjects, the unsupervised learning method was the only one successful at fully removing the artifact. This approach can easily be expanded to other modalities of physiological data in complex settings.

Clinical Relevance— Robust artifact detection methods allow for the use of diverse physiological data even in complex clinical settings to inform diagnostic and therapeutic decisions.

Keywords: Physiological systems modeling - Signal processing in simulation, Neural networks and support vector machines in biosignal processing and classification, Physiological systems modeling - Signal processing in physiological systems
Abstract: Auditory models have been adopted for years to simulate characteristics of the human auditory processing for normal and hearing-impaired listeners. However, individual differences due to varying degrees of frequency-dependent hearing damage hinders the simulation of auditory processing on an individualized basis. Here, with a view on precise auditory profiling, recorded distortion product otoacoustic emission (DPOAE) metrics are used to determine individual parameters of cochlear non-linearity to yield individualized human cochlear models, which can be used as pre-processors for hearing-aid and machine-hearing applications. We test whether individualized cochlear models based on DPOAE measurements can simulate the measured DPOAEs and audiograms of normal-hearing and hearing-impaired listeners. Results showed that cochlear models individualized based on DPOAE-grams measured at low stimulus levels or DPOAE thresholds, yield the smallest simulation errors. Clinical Relevance: The outcomes of this study can improve individualized model predictions of auditory function for mixed hearing pathologies, e.g. cochlear synaptopathy in presence of outer-hair-cell loss, and can enhance future individualized hearing-aid algorithms.

Keywords: Connectivity, Physiological systems modeling - Signal processing in physiological systems, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Children with medically refractory epilepsy (MRE) require resective neurosurgery to achieve seizure freedom, whose success depends on accurate delineation of the epileptogenic zone (EZ). Functional connectivity (FC) can assess the extent of epileptic brain networks since intracranial EEG (icEEG) studies have shown its link to the EZ and predictive value for surgical outcome in these patients. Here, we propose a new noninvasive method based on magnetoencephalography (MEG) and high-density (HD-EEG) data that estimates FC metrics at the source level through an “implantation” of virtual sensors (VSs). We analyzed MEG, HD-EEG, and icEEG data from eight children with MRE who underwent surgery having good outcome and performed source localization (beamformer) on noninvasive data to build VSs at the icEEG electrode locations. We analyzed data with and without Interictal Epileptiform Discharges (IEDs) in different frequency bands, and computed the following FC matrices: Amplitude Envelope Correlation (AEC), Correlation (CORR), and Phase Locking Value (PLV). Each matrix was used to generate a graph using Minimum Spanning Tree (MST), and for each node (i.e., each sensor) we computed four centrality measures: betweenness, closeness, degree, and eigenvector. We tested the reliability of VSs measures with respect to icEEG (regarded as benchmark) via linear correlation, and compared FC values inside vs. outside resection. We observed higher FC inside than outside resection (p<0.05) for AEC [alpha (8-12 Hz), beta (12-30 Hz), and broadband (1-50 Hz)] on data with IEDs and AEC theta (4-8 Hz) on data without IEDs for icEEG, AEC broadband (1-50 Hz) on data without IEDs for MEG-VSs, as well as for all centrality measures of icEEG and MEG/HD-EEG-VSs. Additionally, icEEG and VSs metrics presented high correlation (0.6-0.9, p<0.05). Our data support the notion that the proposed method can potentially replicate the icEEG ability to map the epileptogenic network in children with MRE.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Machine learning methods, such as deep learning, show promising results in the medical domain. However, the lack of interpretability of these algorithms may hinder their applicability to medical decision support systems. This paper studies an interpretable deep learning technique, called SincNet. SincNet is a convolutional neural network that efficiently learns customized band-pass filters through trainable sinc-functions. In this study, we use SincNet to analyze the neural activity of individuals with Autism Spectrum Disorder (ASD), who experience characteristic differences in neural oscillatory activity. In particular, we propose a novel SincNet-based neural network for detecting emotions in ASD patients using EEG signals. The learned filters can be easily inspected to detect which part of the EEG spectrum is used for predicting emotions. We found that our system automatically learns the high-α (9-13 Hz) and β (13-30 Hz) band suppression often present in individuals with ASD. This result is consistent with recent neuroscience studies on emotion recognition, which found an association between these band suppressions and the behavioral deficits observed in individuals with ASD. The improved interpretability of SincNet is achieved without sacrificing performance in emotion recognition.

Keywords: Signal pattern classification, Time-frequency and time-scale analysis - Wavelets, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Motion patterns in newborns contain important information. Motion patterns change upon maturation and changes in the nature of motion may precede critical clinical events such as the onset of sepsis, seizures and apneas. However, in clinical practice, motion monitoring is still limited to observations by caregivers. In this study, we investigated a practical yet reliable method for motion detection using routinely used physiological signals in the patient monitor. Our method calculated motion measures with a continuous wavelet transform (CWT) and a signal instability index (SII) to detect gross-motor motion in 15 newborns using 40 hours of physiological data with annotated videos. We compared the performance of these measures on three signal modalities (electrocardiogram ECG, chest impedance, and photo plethysmography). In addition, we investigated whether their combinations increased performance. The best performance was achieved with the ECG signal with a median (interquartile range, IQR) area under receiver operating curve (AUC) of 0.92(0.87-0.95), but differences were small as both measures had a robust performance on all signal modalities. We then applied the algorithm on combined measures and modalities. The full combination outperformed all single-modal methods with a median (IQR) AUC of 0.95(0.91-0.96) when discriminating gross-motor motion from still. Our study demonstrates the feasibility of gross-motor motion detection method based on only clinically-available vital signs and that best results can be obtained by combining measures and vital signs.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Time-frequency and time-scale analysis - Time-frequency analysis, Principal component analysis
Abstract: Hand gesture decoding is a key component of controlling prosthesis in the area of Brain Computer Interface (BCI). This study is concerned with classification of hand gestures, based on Electrocorticography (ECoG) recordings. Recent studies have utilized the temporal information in ECoG signals for robust hand gesture decoding. In our preliminary analysis on ECoG recordings of hand gestures, we observed different power variations in six frequency bands ranging from 4 to 200 Hz. Therefore, the current trend of including temporal information in the classifier was extended to provide equal importance to power variations in each of these frequency bands. Statistical and Principal Component Analysis (PCA) based feature reduction was implemented for each frequency band separately, and classification was performed with a Long Short-Term Memory (LSTM) based neural network to utilize both temporal and spatial information of each frequency band. The proposed architecture along with each feature reduction method was tested on ECoG recordings of five finger flexions performed by seven subjects from the publicly available `fingerflex' dataset. An average classification accuracy of 82.4% was achieved with the statistical based channel selection method which is an improvement compared to state-of-the-art methods.

Keywords: Neural networks and support vector machines in biosignal processing and classification
Abstract: We propose a computationally efficient algorithm for seizure detection. Instead of using a purely data-driven approach, we develop a hybrid model-based/data-driven method, combining convolutional neural networks with factor graph inference. On the CHB-MIT dataset, we demonstrate that the proposed method can generalize well in a 6 fold leave-4-patientout evaluation. Moreover, it is shown that our algorithm can achieve as much as 5% absolute improvement in performance compared to previous data-driven methods. This is achieved while the computational complexity of the proposed technique is a fraction of the complexity of prior work, making it suitable for real-time seizure detection.

Keywords: Signal pattern classification
Abstract: Emotion calibration is measured by the valence and arousal scales and the ideal center is used to directly divide valence arousal into high scores and low scores. This division method has a big classification and labeling defect, and the influence of emotion stimulation material on the subjects cannot be accurately measured. To address this problem, this paper proposes an EEG emotion recognition algorithm (DW-FBCSP: Distance Weighted Filter Bank Common Spatial Pattern) based on scale distance weighted optimization to optimize the classification according to the distance of the scores from ideal center. This method is a natural extension of CSP that optimize the user's EEG signal projection matrix. Then, the LDA classifier is used to recognize emotions using the features set which fused the selected features and the features extracted by the projection matrix. The results show that the mean correct rate of the valence and arousal achieves 81.14% and 84.45% using the DEAP dataset. The results demonstrate that our proposed method outperforms better than some other results published in recent years.

Keywords: Principal and independent component analysis - Blind source separation, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Fetal heart rate monitoring using the abdominal electrocardiograph (ECG) is an important topic for the diagnosis of heart defects. Many studies on fetal heart rate detection have been presented, however, their accuracy is still unsatisfactory. That is because the fetal ECG waveform is contaminated by maternal ECG interference, muscle contractions, and motion artifacts. One of the conventional methods is to detect the R-peaks from the integrated power of the frequency corresponding to the fetal heartbeats. However, the detection accuracy of the R-peaks is not enough. In this paper, we propose a method to generate the candidates of R-peaks using the first derivative of the signal and to pick up the estimated heartbeats by a multiple weighting function. The proposed multiple weighting function is designed by the Gaussian distribution, of which parameters are set from a grid search with the goal of minimizing the standard deviation of RR intervals (neighboring R-peaks intervals). The validation for the proposed framework has been evaluated on real-world data, which got the better accuracy than the conventional method that detects R-peaks from the integrated power and uses the weighting function produced by a fixed parameter of Gaussian distribution. The averaged absolute error (AAE) which compares the estimated fetal heart rate and the reference fetal heart rate has been decreased by 17.528 bpm.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Physiological systems modeling - Systems identification
Abstract: In this contribution, we focused on optimising a dynamic flow-based shear stress system to achieve a reliable platform for cell shear modulus (stiffness) and viscosity assessment using quantitative phase imaging. The estimation of cell viscoelastic properties is influenced by distortion of the shear stress waveform, which is caused by the properties of the flow system components (i.e., syringe, flow chamber and tubing). We observed that these components have a significant influence on the measured cell viscoelastic characteristics. To suppress this effect, we applied a correction method utilizing parametric deconvolution of the flow system's optimized impulse response. Achieved results were compared with the direct fitting of the Kelvin-Voigt viscoelastic model and the basic steady-state model. The results showed that our novel parametric deconvolution approach is more robust and provides a more reliable estimation of viscosity with respect to changes in the syringe's compliance compared to Kelvin-Voigt model.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods, Signal pattern classification
Abstract: This study attempted to classify a small amount of electroencephalogram (EEG) data on five states: four tasks involving right index-finger flexion (kinesthetic motor imagery, visual motor imagery, motor execution, and motor observation) and resting with eyes open. We employed a convolutional neural network (CNN) as a classifier and compared the classification accuracies of two types of CNNs: 1) a "single five-class CNN," which classified the aforementioned states with a single CNN and 2) "multiple two-class CNNs," wherein ten CNNs that classify pairs of states were combined. In addition, the classification accuracies were compared between two scenarios: one wherein the EEGs from all 19 scalp probe electrodes (19-channel EEG) were adopted as input data for the CNN, and the other wherein the EEGs of four regions closely related to the motor execution and observation of the index finger (4-channel EEG) were adopted. The classification accuracies of the single five-class CNN with 19- and 4-channel EEGs were 48.2 ± 5.9% and 46.6 ± 6.9%, respectively, and those of the multiple two-class CNNs with 19- and 4-channel EEGs were 52.8 ± 9.7% and 47.5 ± 9.4%, respectively. These results indicate the effectiveness of multiple two-class CNNs that utilize the EEGs of all scalp electrodes as input data for classifying motor imagery, execution, and observation, even in the case of the marginal dataset.

Keywords: Neural networks and support vector machines in biosignal processing and classification
Abstract: Private spaces like apartments and vehicles are not yet fully exploited for health monitoring, which includes continuous measurement of biosignals. This work proposes sensor fusion for robust heartbeat detection in the noisy and dynamic driving environment. We use four sensors: electrocardiography (ECG), ballistocardiography (BCG), photoplethysmography (PPG), and image-based PPG (iPPG). As ground truth, we record a 3-lead ECG with wet electrodes attached to the chest. Twelve healthy volunteers are monitored in rest and during driving, each for 11 min. We propose sensor fusion using convolutional neural networks to detect the sensor combination delivering the most accurate heart rate measurement. For rest, we achieve scores of 95.16% (BCG + iPPG), 96.08% (ECG + iPPG), 96.35% (ECG + BCG), 96.53% (ECG + PPG), 96.58% (PPG + iPPG), and 97.15% (BCG + PPG). In motion, the highest scores are 92.46% (BCG + iPPG, PPG + iPPG, ECG + iPPG), 92.83% (ECG + PPG), 93.03% (BCG + PPG), and 93.08% (ECG + BCG). Fusing all four signals with the best fusion approach results in scores of 97.24% (rest) and 94.38% (motion). We conclude that sensor fusion allows robust heartbeat measurement of car drivers to support continuous and unobtrusive health monitoring for early disease detection.

Keywords: Connectivity, Principal component analysis, Signal pattern classification
Abstract: Many patients with mental illnesses character-ized by impaired cognitive control have no relief from gold-standard clinical treatments resulting in a pressing need for new alternatives. This paper develops a neural decoder to detect task engagement in ten human subjects during a conflict-based behavioral task known as the multi-source interference task (MSIT). Task engagement is of particular interest here because closed-loop brain stimulation during those states can augment decision-making. The functional connectivity patterns of the electrodes are extracted. A principal component analysis of these patterns is carried out and the ranked principal components are used as inputs to train subject-specific linear support vector machine classifiers. In this paper, we show that task engagement can be differentiated from background brain activity with a median accuracy of 89.7%. This was accom-plished by constructing distributed functional networks from local field potentials recording during the task performance. A further challenge is that goal-directed efforts take place over higher temporal resolution. Task engagement must thus be detected at a similar rate for proactive intervention. We show that our algorithms can detect task engagement from neural recordings in less than 2 seconds; this can be further improved using an application-specific device.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Biosignal classification
Abstract: Electrocardiogram (ECG) is mainly used by medical domain to diagnose arrhythmia. With the development of deep learning algorithms in the ECG classification field, related algorithms have achieved very high accuracy. However, the training of deep learning algorithms always requires large amounts of samples, while the labeled samples are often lacked in the field of medical signals. Therefore, the performance of deep learning algorithms will be greatly restricted. To overcome the sample scarcity problem, we propose a few-shot ECG classification approach based on the Siamese network. This network architecture first uses two one-dimensional convolutional neural network (CNN) that share weights to extract feature vectors of the paired input signals. Then, L1-distance between the two feature vectors is calculated and inputted into the fully connected layer with an activation function sigmoid to determine whether the input pairs belong to same category. We validated our method on the MIT-BIH arrhythmia database. By experiments, our method performs better than existing networks under the circumstance of extremely few amount of data.

Keywords: Connectivity, Nonlinear dynamic analysis - Phase locking estimation, Nonlinear dynamic analysis - Biomedical signals
Abstract: Phonemes are classified into different phonological categories based on the place and manner of articulation. We investigate the differences in the neural correlates of imagination of prompts from two phonological categories, namely nasal and bilabial consonants. Mean phase coherence is used as a metric for measuring the phase synchronisation between different electrode pairs in six cortical regions (auditory, motor, prefrontal, sensorimotor, somatosensory and premotor) during the imagery of nasal and bilabial consonants. Statistically significant difference at 95% confidence interval is observed in beta and lower-gamma bands in various cortical regions. Our observations are inline with DIVA (directions into velocities of articulators) model and dual stream prediction model and support the hypothesis that phonological categories not only exist in articulated speech but can also be distinguished from the EEG of imagined speech.

Clinical relevance— Identification of neural correlates of imagined speech helps in developing better prompts for imagined speech based brain-computer interfaces (BCI) leading to improvements in both accuracy and degrees of freedom. BCIs play a significant role as technology aids for differently abled individuals and for patients with disorders of consciousness (DoC). It also helps in better understanding the neural correlates of psychotic disorders such as schizophrenia and paranoia where auditory hallucinations is a major symptom.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Blood pressure (BP) is an important indicator for prevention and management of cardiovascular diseases. Alongside the improvement in sensors and wearables, photoplethysmography (PPG) appears to be a promising technology for continuous, non-invasive and cuffless BP monitoring. Previous attempts mainly focused on features extracted from the pulse morphology. In this paper, we propose to remove the feature engineering step and automatically generate features from an ensemble average (EA) PPG pulse and its derivatives, using convolutional neural network and a calibration measurement. We used the large VitalDB dataset to accurately evaluate the generalization capability of the proposed model. The model achieved mean errors of -0.24 ± 11.56 mmHg for SBP and -0.5 ± 6.52 mmHg for DBP. We observed a considerable reduction in error standard deviation of above 40% compared to the control case, which assumes no BP variation. Altogether, these results highlight the capability to model the dependency between PPG and BP.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods, Signal pattern classification
Abstract: Myocardial ischemia, consisting in a reduction of blood flow to the heart, may cause sudden cardiac death by myocardial infarction or trigger serious abnormal rhythms. Thus, its timely identification is crucial. The Repeated Structuring and Learning Procedure (RS&LP), an innovative constructive algorithm able to dynamically create neural networks (NN) alternating structuring and learning phases, was previously found potentially useful for myocardial ischemia detection. However, performance of created NN depends on three parameters, the values of which need to be set a priori by the user: maximal number of layers (NL), maximal number of initializations (NI) and maximal number of confirmations (NC). A robustness analysis of RS&LP to varying values of NL, NI and NC is fundamental for clinical applications concerning myocardial ischemia detection but was never performed before; thus, it was the aim the present study. Thirteen serial ECG features were extracted by pairs of ECGs belonging to 84 cases (patients with induced myocardial ischemia) and 398 controls (patients with no myocardial ischemia) and used as inputs to learn (50% of population) and test (50% of population) NNs with varying values of NL (1,2,3,4,10), NI (50,250,500,1000,1500) and NC (2,5,10,20,50). Performance of obtained NNs was compared in terms of area under the curve (AUC) of the receiver operating characteristics. Overall, 13 NNs were considered; 12 (92%) were characterized by AUC≥80% and 4 (31%) by AUC≥85%. Thus, RS&LP proved to be robust when creating NNs for detecting of myocardial ischemia.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Data mining and big data methods - Biosignal classification, Nonlinear dynamic analysis - Biomedical signals
Abstract: Seizures represent one of the most challenging issues of the neonatal period’s neurological emergency. Due to the heterogeneity of etiologies and clinical characteristics, seizures recognition is tricky and time-consuming. Currently, the gold standard for seizure diagnosis is Electroencephalography (EEG), whose correct interpretation requires a highly specialized team. Thus, to speed up and facilitate the detection of ictal events, several EEG-based Neonatal Seizure Detectors (NSDs) have been proposed in the literature. Research is currently exploiting more simple and less invasive approaches, such as Electrocardiography (ECG). This work aims at developing an ECG-based NSD using a Generalized Linear Model with features extracted from Heart Rate Variability (HRV) measures as input. The method is validated on a public dataset of 52 subjects (33 with seizures and 19 seizure-free). Achieved encouraging results show 69% Concatenated Area Under the ROC Curve (AUCcc) for the automatic detection of windows with seizure events, confirming that HRV features can be useful to catch the cardio-regulatory system alterations due to neonatal seizure events, particularly those related to Hypoxic-Ischaemic Encephalopathies. Thus, results suggest the use of ECG-based NSDs in clinical practice, especially when a timely diagnosis is needed and EEG technologies are not readily available.

Keywords: Connectivity, Multivariate methods, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Over the past twenty years, functional connectivity of the human brain has been studied in detail using tools from complex network theory. These methods include graph theoretic metrics ranging from the micro-scale such as the degree of a node to the macro-scale such as the small worldness of the brain network. However, most of these network models focus on average activity within a time window of interest and given frequency band. Therefore, they cannot capture the changes in network connectivity across time and different frequency bands. Recently, multilayer brain networks have attracted a lot of attention as they can capture the full view of neuronal connectivity. In this paper, we introduce a multilayer view of the functional connectivity network of the brain, where each layer corresponds to a different frequency band. We construct multi-frequency connectivity networks from electroencephalogram data where the intra-layer edges are quantified by phase synchrony while the inter-layer edges are quantified by phase-amplitude coupling. We then introduce multilayer degree, participation coefficient and clustering coefficient to quantify the centrality of nodes across frequency layers and to identify the importance of different frequency bands. The proposed framework is applied to electroencephalogram data collected during a study of error monitoring in the human brain.

Keywords: Coupling and synchronization - Coherence in biomedical signal processing
Abstract: During minimally invasive and image-guided procedures, vital parameters have to be recorded for patient safety. In the MRT environment the MHD effect emerges, under the high magnetic field strength in the signal recording of an ECG system. To allow the investigation of this effect, newly developed wireless ECG and ICG devices using a SNTP server for accurate synchronization of collected data will be presented. All devices which are part of the network are based on the Espressif ESP32, a dual core MCU which integrates simultaneous Bluetooth and Wi-Fi communications. ECG signal collected at 1000 samples per second and ICG signal collected at 250 samples per second are visualized by a software interface running on an Operator Workstation, which additionally sets the SNTP server and the wireless access point. Moreover, a reusable custom set of electrodes for tetrapolar ICG with flexible electronics manufacturing techniques is developed. According to measurement results the ICG device and electrodes comply with the IEC 60601-1 standard. If needed, the system architecture can be scaled with the inclusion of more custom-built wireless devices (e.g. pulse oximeters, pressure meters). Its affordability would make it a feasible technological platform for establishing multicentric studies, data collected from subject groups would conform to the same standards of quality and security and be made easily available to the scientific community. In future work, the system will be made compliant for use in an MRT environment where time synchronous data sampling of ECG and ICG would significantly enhance the study of the MHD effect generated by vessel blood flow.

Keywords: Principal and independent component analysis - Blind source separation, Independent component analysis, Principal component analysis
Abstract: Abstract— The mismatch response (MMR) is thought to be a neurophysiological measure of novel auditory detection that could serve as a translational biomarker of various neurological diseases. When recorded with electroencephalography (EEG) or magnetoencephalography (MEG), the MMR is traditionally extracted by subtracting the event-related potential/field (ERP/ERF) elicited in response to “deviant” sounds that occur randomly within a train of repetitive “standard” sounds. To overcome the limitations of this subtraction procedure, we propose a novel method which we call weighted-BSST/k, which uses only the deviant response to derive the MMR. We hypothesized that this novel weighted-BSST/k method highlights responses related to the detection of the deviant stimulus and is more sensitive than independent component analysis (ICA). To test this hypothesis and the validity and efficacy of the weighted-BSST/k in comparison with ICA (infomax), we evaluated the methods in 12 healthy adults. Auditory stimuli were presented at a constant rate of 2 Hz. Frequency MMRs at a sensor level were obtained from the bilateral temporal lobes with the subtraction approach at 96–276 ms (the MMR time range), defined on the basis of spatio-temporal cluster permutation analysis. In the application of the weighted-BSST/k, the deviant responses were given a constant weight on the MMR time range. The ERF elicited by the weighted deviant responses demonstrated one or a few dominant components representing the MMR with a high signal-to-noise ratio and similar topography to that of the sensor space analysis using the conventional subtraction approach. In contrast, infomax or weighted-infomax revealed many minor or pseudo components as constituents of the MMR. Our new approach may assist in using the MMR in basic and clinical research. Clinical Relevance—Our proposed method opens a new and potentially useful way to analyze event-related MEG/EEG data.

Keywords: Data mining and big data methods - Biosignal classification
Abstract: Circadian rhythms of blood pressure (BP) have key diagnostic significance in the assessment of hypertension. The night-time dip or rise in BP (10-20% decrease or increase compared to daytime BP), for example, has been shown to be a strong indicator for cardiovascular disease. However, current methods for assessing the circadian rhythm of BP can be disruptive to sleep, work, and daily activities. Pulse arrival time (PAT) has been suggested as a surrogate measure of BP. This work investigates the presence of a circadian rhythm in PAT and evaluates its application to classify nocturnal BP dip or rise. 769 patients who were discharged home from the ICU were selected from the MIMIC database. Our results show a clear and observable circadian rhythm of PAT that is strongly inversely correlated with BP (R = -0.89). The ratios between nocturnal and diurnal changes in PAT accurately classifies an individual as a nocturnal BP dipper (AUC = 0.72) or a riser (AUC = 0.71).

Keywords: Nonlinear dynamic analysis - Biomedical signals, Physiological systems modeling - Multivariate signal processing
Abstract: Photoplethysmography (PPG) is a completely noninvasive, optical method of assessing blood flow dynamics in peripheral vasculature. Wearable devices for PPG recording are becoming increasingly popular, due to their cost-effectiveness and ease of use. For these reasons, many recent scientific studies have proposed the use of pulse rate variability (PRV) extracted from PPG as a surrogate for heart rate variability (HRV), in monitoring autonomic activity and cardiovascular health. In this work, we used a cross-mapping approach, a methodology based on chaos theory, to compare PRV and HRV dynamics, and investigate their agreement according to age and gender of healthy subjects. We used ECG and PPG data acquired from 57 subjects (41 young and 16 elderly) during resting state in the supine position. Signals were gathered from the publicly available VORTAL dataset. Our results showed a statistically significant decrease of PRV reliability as an HRV surrogate in old participants, which was confirmed as significant when only men subjects were analyzed (p-value<0.01). Our findings, although preliminary, suggest greater caution in the use of PPG devices for monitoring cardiovascular health, especially in elderly men.

Keywords: Data mining and big data methods - Biosignal classification, Data mining and big data methods - Pattern recognition, Signal pattern classification
Abstract: Recent studies have attempted to recognize emotions by extracting features from electroencephalographic (EEG) signals using either linear and stationary, or linear and non-stationary transformations. However, as EEG signals are non-linear and non-stationary, it seems that a non-linear and non-stationary transformation may be more suitable. Despite the attractiveness of this hypothesis, until now, little studies have used such transformation. The current work presents a comparison between an approach to recognize positive and negative emotions using a non-linear and non-stationary transformation (Hilbert-Huang Transformation) with an approach using linear and non-stationary transformation (Discrete Wavelet Transform). The two approaches were compared using 200 EEG signals recorded from 10 subjects. The comparison indicated that the HHT-based approach statistically significantly classified emotions more accurately than the Wavelet-based approach (P < 0.02). This result implies that HHT is a promising transformation to increase the prediction of emotional states, thereby helping to designing and developing more robust emotion recognition approaches.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Independent component analysis
Abstract: EEG can be used to characterize the electrical activity of the cerebral cortex, but it is also susceptible to interference. Compared with the other artifacts, Electrooculogram (EOG) artifacts have a greater impact on EEG processing and are more difficult to remove. Here, we mainly compared the regression and ICA algorithms both based on the EOG channels for the effect of removing EOG artifacts in the Stroop experiment of methamphetamine addicts. From the perspective of time domain and power spectral density, the ICA algorithm based on the EOG channels is more effective. However, the regression algorithm based on the EOG channels is less complex, more time-saving, and more suitable for real-time tasks.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Neural networks and support vector machines in biosignal processing and classification
Abstract: The paper proposes accurate Blood Pressure Monitoring (BPM) based on a single-site Photoplethysmographic (PPG) sensor and provides an energy-efficient solution on edge cuffless wearable devices. Continuous PPG signal preprocessed and used as input of the Artificial Neural Network (ANN), and outputs systolic BP (SBP), diastolic BP (DBP), and mean arterial BP (MAP) values for each heartbeat. The improvement of the BPM accuracy is obtained by removing outliers in the preprocessing step and the whole-based inputs compared to parameter-based inputs extracted from the PPG signal. Performance obtained is 3.42 ± 5.42 mmHg (MAE ± RMSD) for SBP, 1.92 ± 3.29 mmHg for DBP, and 2.21 ± 3.50 mmHg for MAP which is competitive compared to other studies. This is the first BPM solution with edge computing artificial intelligence as we have learned so far. Evaluation experiments on real hardware show that the solution takes 42.2 ms, 18.2 KB RAM, and 2.1 mJ average energy per reading.

Keywords: Nonlinear dynamic analysis - Biomedical signals, Principal component analysis
Abstract: Repetitive atrial conduction patterns are often observed during human atrial fibrillation (AF). Repetitive patterns may be associated with AF drivers such as focal and micro-reentrant mechanisms. Therefore, tools for repetitive activity detection are of great interest as they may allow to identify the leading electrophysiological AF mechanism in an individual patient. Recurrence plots (RP) are efficient tools for repetitive activity visualization. Construction of an RP requires embedding of epicardial atrial electrograms into a phase space. In this study, we compared the conventional Takens’ embedding approach and three novel approaches based on a priori AF cycle length (AFCL) information. Approaches were bench-marked based on the similarity of the RPs they produce with a previously proposed technique, the sensitivity and specificity to detect the repetitive patterns, as well as the capability to estimate overall repetitive pattern prevalence. All techniques were applied to high-density epicardial direct contact mapping recordings in AF patients with paroxysmal AF (n=12) and persistent AF (n=9). Compared to a reference method the proposed novel techniques achieved significantly higher similarity and sensitivity values (p<0.01) than conventional embedding, in both paroxysmal and persistent patients. Moreover, estimated prevalences were robust against the various degrees of AF complexity present in the recordings.

Keywords: Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis, Nonlinear dynamic analysis - Biomedical signals, Time-frequency and time-scale analysis - Wavelets
Abstract: Mechanically ventilated patients typically exhibit abnormal respiratory sounds. Squawks are short inspiratory adventitious sounds that may occur in patients with pneumonia, such as COVID-19 patients. In this work we devised a method for squawk detection in mechanically ventilated patients by developing algorithms for respiratory cycle estimation, squawk candidate identification, feature extraction, and clustering. The best classifier reached an F1 of 0.48 at the sound file level and an F1 of 0.66 at the recording session level. These preliminary results are promising, as they were obtained in noisy environments. This method will give health professionals a new feature to assess the potential deterioration of critically ill patients.

Keywords: Time-frequency and time-scale analysis - Nonstationary analysis and modeling, Physiological systems modeling - Signal processing in physiological systems
Abstract: The increasing population size of the elderly is fostering the development of telehealth and assisted living systems. In this respect, monitoring vital biophysical conditions using wireless devices, such as the wireless electrocardiogram (WECG), plays a pivotal role in telemonitoring. However, the freedom of movement brings with it motion artifacts, the magnitude of which can be significant enough to interfere with the cardiac signals. To reason about and remove the artifacts, reference models (signals) are needed. IN the context of WECGs, one way to construct these models is to employ motion sensors that can pick up the motion affecting the electrodes of the WECGs. In this paper, we experimentally examine the spectra of motion artifacts and the existence of correlations between inertial sensors and motion artifacts. We make use of three different types of sensors (3D accelerometer, 3D gyroscope, and skin-electrode impedance sensor) to assess the characteristics of different movement types. We found that the spectra of motion artifacts are determined by the type of movement. While lower-intensity motion artifacts (e.g., bending forward) are most pronounced below 2 Hz, others (e.g., running) manifest themselves in a series of distinct peaks between 1–10 Hz.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Nonlinear dynamic analysis - Biomedical signals, Physiological systems modeling - Signal processing in physiological systems
Abstract: Recently, deep learning algorithms have been used widely in emotion recognition applications. However, it is difficult to detect human emotions in real-time due to constraints imposed by computing power and convergence latency. This paper proposes a real-time affective computing platform that integrates an AI System-on-Chip (SoC) design and multimodal signal processing systems composed of electroencephalogram (EEG), electrocardiogram (ECG), and photoplethysmogram (PPG) signals. To extract the emotional features of the EEG, ECG, and PPG signals, we used a short-time Fourier transform (STFT) for the EEG signal and direct extraction using the raw signals for the ECG and PPG signals. The long-term recurrent convolution networks (LRCN) classifier was implemented in an AI SoC design and divided emotions into three classes: happy, angry, and sad. The proposed LRCN classifier reached an average accuracy of 77.41% for cross-subject validation. The platform consists of wearable physiological sensors and multimodal signal processors integrated with the LRCN SoC design. The area of the core and total power consumption of the LRCN chip was 1.13 x 1.14 mm^2 and 48.24 mW, respectively. The on-chip training processing time and real-time classification processing time are 5.5 us and 1.9 us per sample. The proposed platform displays the classification results of emotion calculation on the graphical user interface (GUI) every one second for real-time emotion monitoring.

Keywords: Signal pattern classification, Physiological systems modeling - Signal processing in physiological systems, Neural networks and support vector machines in biosignal processing and classification
Abstract: Auscultation of respiratory sounds is the primary tool for screening and diagnosing lung diseases. Automated analysis, coupled with digital stethoscopes, can play a crucial role in enabling tele-screening of fatal lung diseases. Deep neural networks (DNNs) have shown potential to solve such problems, and are an obvious choice. However, DNNs are data hungry, and the largest respiratory dataset ICBHI has only 6898 breathing cycles, which is quite small for training a satisfactory DNN model. In this work, RespireNet, we propose a simple CNN-based model, along with a suite of novel techniques---device specific fine-tuning, concatenation-based augmentation, blank region clipping, and smart padding---enabling us to efficiently use the small-sized dataset. We perform extensive evaluation on the ICBHI dataset, and improve upon the state-of-the-art results for 4-class classification by 2.2%

Keywords: Connectivity, Multivariate methods
Abstract: Comprehension of speech in noise is a challenge for hearing-impaired (HI) individuals. Electroencephalography (EEG) provides a tool to investigate the effect of different levels of signal-to-noise ratio (SNR) of speech. Most studies with EEG have focused on spectral power in well-defined frequency bands such as alpha band. In this study, we investigate how local functional connectivity, i.e. functional connectivity within a localized region of the brain, is affected by two levels of SNR. Twenty-two HI participants performed a continuous speech in noise task at two different SNRs (+3 dB and +8 dB). The local connectivity within eight regions of interest was computed by using a multivariate phase synchrony measure on EEG data. The results showed that phase synchrony increased in the parietal and frontal area as a response to increasing SNR. We contend that local connectivity measures can be used to discriminate between speech-evoked EEG responses at different SNRs.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Multivariate methods, Signal pattern classification
Abstract: Continuous and multimodal stress detection has been performed recently through wearable devices and machine learning algorithms. However, a well-known and important challenge of working on physiological signals recorded by conventional monitoring devices is missing data due to sensors insufficient contact and interference by other equipment. This challenge becomes more problematic when the user/patient is mentally or physically active or stressed because of more frequent conscious or subconscious movements. In this paper, we propose ReLearn, a robust machine learning framework for stress detection from biomarkers extracted from multimodal physiological signals. ReLearn effectively copes with missing data and outliers both at training and inference phases. ReLearn, composed of machine learning models for feature selection, outlier detection, data imputation, and classification, allows us to classify all samples, including those with missing values at inference. In particular, according to our experiments and stress database, while by discarding all missing data, as a simplistic yet common approach, no prediction can be made for 34% of the data at inference, our approach can achieve accurate predictions, as high as 78%, for missing samples. Also, our experiments show that the proposed framework obtains a cross-validation accuracy of 86.8% even if more than 50% of samples within the features are missing.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Nonlinear dynamic analysis - Biomedical signals, Data mining and big data methods - Biosignal classification
Abstract: This paper proposes a novel lightweight method using the multitaper power spectrum to estimate arousal levels at wearable devices. We show that the spectral slope (1/f) of the electrophysiological power spectrum reflects the scale-free neural activity. To evaluate the proposed feature's performance, we used scalp EEG recorded during anesthesia and sleep with technician-scored Hypnogram annotations. It is shown that the proposed methodology discriminates wakefulness from reduced arousal solely based on the neurophysiological brain state with more than 80% accuracy. Therefore, our findings describe a common electrophysiological marker that tracks reduced arousal states, which can be applied to different applications (e.g., emotion detection, driver drowsiness). Evaluation on hardware shows that the proposed methodology can be implemented for devices with a minimum RAM of 512 KB with 55 mJ average energy consumption.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification
Abstract: Intracranial pressure (ICP) pulse waveform, i.e., the shape of the ICP signal over a single cardiac cycle, is regarded as a potential source of information about intracranial compliance. In this study we aimed to compare the results of automatic classification of ICP pulse shapes on a scale from normal to pathological with other ICP pulse-derived metrics. Additionally, identification of artifacts was performed simultaneously with pulse classification to assess the effect of artifact removal on the results. Data from 35 traumatic brain injury (TBI) patients were analyzed retrospectively in terms of dominant waveform shape, mean ICP, mean amplitude of ICP (AmpICP), mean index of compensatory reserve (RAP index), and their association with the patient’s clinical outcome. Our results show that patients with poor outcome exhibit more pathological waveform shape than patients with good outcome. More pathological ICP pulse shape is associated with higher mean ICP, mean AmpICP, and RAP.

Clinical relevance— In the clinical setting, ICP pulse waveform analysis could potentially be used to complement the commonly monitored mean ICP and improve the assessment of intracranial compliance in TBI patients. Artifact removal from the ICP signal could reduce the frequency of false positive detection of clinically adverse events.

Keywords: Signal pattern classification - Genetic algorithms, Data mining and big data methods - Pattern recognition, Physiological systems modeling - Signal processing in physiological systems
Abstract: This paper focuses on a new algorithm for solving optimization problems using the nature of food search behaviour of caterpillars. The paper describes how the periscopic, pheromonic and fractal search properties analogous to the caterpillars, can aid in designing a new optimization algorithm. The performance characteristics of the new method is compared using 26 standard test functions and the area under the curve of the fitness evaluations is used to validate and compare the proposed algorithms against existing related works. The proposed algorithm is found to be efficient when compared with the existing methods. The proposed algorithm is then tested on a real world problem to remove signal noise from eye gaze data, effectively.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods, Signal pattern classification
Abstract: With the purpose of providing an external human-machine interaction platform for the elderly in need, a novel facial surface electromyography based silent speech recognition system was developed. In this study, we propose a deep learning architecture named Parallel-Inception Convolutional Neural Network (PICNN), and employ up-to-date feature extraction method log Mel frequency spectral coefficients (MFSC). To better meet the requirements of our target users, a 100-class dataset containing daily life-related demands was designed and generated for the comparative experiments. According to experimental results, the highest recognition accuracy of 88.44% was achieved by proposed recognition framework based on MFSC and PICNN, exceeding the performance of state-of-the-art deep learning algorithms such as CNN, VGGNet and Inception CNN (3.22%, 4.09% and 1.19%, respectively). These findings suggest that the newly developed silent speech approach holds promise to provide a more reliable communication channel, and the application scenery of speech recognition technology has been expanded at the same time.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Studies have suggested that the hippocampus (Hp) plays an important role in spatial learning and avian Hp is thought to have similar functions with mammals. However, the dynamic neural pattern of hippocampal activity is still unclear in the continuous spatial learning processes of birds. In this study, we recorded the behavioral data and local field potential (LFP) activity from Hp of pigeons performing goal-directed behavior. Then the spectral properties and time-frequency properties of the LFPs are analyzed, comparing with the behavioral changes during spatial learning. The results indicated that the power of the LFP signal in the gamma band shown decreasing trend during spatial learning. Time-frequency analysis results shown that the hippocampal gamma activity was weakened along with the learning process. The results indicate that spatial learning correlated with the decreased gamma activity in Hp and hippocampal neural patterns of pigeons were modulated by goal-directed behavior.

Keywords: Physiological systems modeling - Signal processing in physiological systems
Abstract: The electrooculography (EOG) signal baseline is subject to drifting, and several different techniques to mitigate this drift have been proposed in the literature. Some of these techniques, however, disrupt the overall ocular pose-induced DC characteristics of the EOG signal and may also require the data to be zero-centred, which means that the average point of gaze (POG) has to lie at the primary gaze position. In this work, we propose an alternative baseline drift mitigation technique which may be used to de-drift EOG data collected through protocols where the subject gazes at known targets. Specifically, it uses the target gaze angles (GAs) in a battery model of the eye to estimate the ocular pose-induced component, which is then used for baseline drift estimation. This method retains the overall signal morphology and may be applied to non-zero-centred data. The performance of the proposed baseline drift mitigation technique is compared to that of five other techniques which are commonly used in the literature, with results showing the general superior performance of the proposed technique.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: Cardiovascular diseases are the number one cause of death worldwide. Detecting cardiovascular diseases in its early stages could effectively reduce the mortality rate by providing timely treatment. In this study, we propose a new methodology to detect arrythmias, using 2D Convolutional Neural Networks. The main characteristic of the proposed methodology is the use of 15 x15 pixels gray-level images, containing the values of a heartbeat of the ECG signal. This work aims to detect 17 arrythmias. To validate and test the proposed methodology, MIT-BIH database, the main benchmark database available in literature, was used. When compared to other results previously published, the obtained precision, 92.31%, is in the state-of-the-art.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Emotion recognition based on electroencephalography (EEG) signals has been receiving significant attention in the domains of affective computing and brain-computer interfaces (BCI). Although several deep learning methods have been proposed dealing with the emotion recognition task, developing methods that effectively extract and use discriminative features is still a challenge. In this work, we propose the novel spatio-temporal attention neural network (STANN) to extract discriminative spatial and temporal features of EEG signals by a parallel structure of multi-column convolutional neural network and attention-based bidirectional long-short term memory. Moreover, we explore the inter-channel relationships of EEG signals via graph signal processing (GSP) tools. Our experimental analysis demonstrates that the proposed network improves the state-of-the-art results in subject-wise, binary classification of valence and arousal levels as well as four-class classification in the valence-arousal emotion space when raw EEG signals or their graph representations, in an architecture coined as GFT-STANN, are used as model inputs.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: The prefrontal asymmetry (FA) in the alpha band is a well-known physiological correlate of the emotional valence. Several methods for assessing the FA have been proposed in literature, but no studies have compared their effectiveness in a comprehensive way. In this study we first investigated whether the association between FA and valence depends on the computational methods and then, we identified the best one, namely the one giving the highest correlation with the self-reports. The investigated factors were the presence of a normalization factor, the computation in time or frequency domain and the cluster of electrodes used. All the analyses were implemented on the validated DEAP dataset. We found that the number and position of the electrodes do not influence the FA, in contrast with both the power computation method and the normalization. By using a spectrogram-based approach and by adding a normalization factor, a correlation of 0.36 between the FA and the self-reported valence was obtained.

Keywords: Independent component analysis, Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis, Physiological systems modeling - Signal processing in physiological systems
Abstract: With the increase in life expectancy, as well as in the performance and complexity of healthcare systems, the need for fast and accurate information has also grown. EEG devices have become more accessible and necessary in clinical practice. In daily activity, artifacts are ubiquitous in EEG signals. They arise from: environmental, experimental and physiological factors, degrade signal quality and render the affected part of the signal useless. This paper proposes an artifact cleaning pipeline including filters and algorithms to streamline the analysis process. Moreover, to better characterize and discriminate artifacts from useful EEG data, additional physiological signals and video data are used, which are correlated with subject’s behavior. We quantify the performance reached by Peak Signal-to-Noise Ratio and clinical visual inspection. The entire research and data collection took place in the laboratories of XPERI Corporation.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Signal pattern classification
Abstract: Virtual reality (VR) technology offers an exciting way to emulate real-life walking conditions that may better elicit changes in emotional state. We aimed to determine whether VR technology is a feasible way to elicit changes in state anxiety during walking. Electrocardiogram data were collected for 18 older adult women while they navigated a baseline walking task, a dual walking task, and four walking VR environments. Using heart rate variability (HRV) analysis, we found that all four of the VR environments successfully elicited a significantly higher level of state anxiety as compared to the walking baseline, with 84% of participants eliciting a significantly lower HRV in each of the four VR conditions as compared to baseline walking. VR was also found to be a more reliable tool for increasing state anxiety as compared to a dual task, where only 47% of participants demonstrated a significantly lower HRV as compared to baseline walking. VR, therefore, could be promising as a tool to elicit changes in state anxiety and less limited in its ability to elicit changes as compared to a traditional dual task condition.

Keywords: Coupling and synchronization - Nonlinear coupling, Neural networks and support vector machines in biosignal processing and classification, Nonlinear dynamic analysis - Biomedical signals
Abstract: Presurgical localization from interictal electrocorticogram (ECoG) and resection of seizure onset zone (SOZ) are difficult processes to achieve seizure freedom. Recently, high frequency oscillations (HFOs) have been recognized as reliable biomarkers for epilepsy surgery which has a relation with the phase of low frequency activities in ECoG. Considering the recent valid biomarker for epilepsy surgery, we hypothesize that the approach of coupling between HFOs and low frequency phases differs SOZ from non-seizure onset zone (NSOZ). This study proposes phase-amplitude coupling (PAC) method to identify SOZ by measuring whether the amplitude of HFOs is coupled with a phase at 2–34 Hz in ECoG. Besides, three machine learning models for PAC-based features are designed for SOZ detection. Four patients with focal cortical dysplasia (FCD) are examined to observe efficiency. Experimental results indicate that the mode of coupling is a potential feature to detect SOZ.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification, Physiological systems modeling - Multivariate signal processing
Abstract: Electrocardiogram (ECG) signals convey immense information that, when properly processed, can be used to diagnose various health conditions including arrhythmia and heart failure. Deep learning algorithms have been successfully applied to medical diagnosis, but existing methods heavily rely on abundant high-quality annotations which are expensive. Self-supervised learning (SSL) circumvents this annotation cost by pre-training deep neural networks (DNNs) on auxiliary tasks that do not require manual annotation. Despite its imminent need, SSL applications to ECG classification remain under-explored. In this work, we propose an SSL algorithm based on ECG delineation and show its effectiveness for arrhythmia classification. Our experiments demonstrate not only how the proposed algorithm enhances the DNN's performance across various datasets and fractions of labeled data, but also how features learnt via pre-training on one dataset can be transferred when fine-tuned on a different dataset.

Keywords: Physiological systems modeling - Signal processing in simulation
Abstract: In recent years the introduction of 5G networks is causing a drastically change of human exposure levels in the radio frequency range. The aim of this paper is on expanding the knowledge on this issue, assessing the exposure levels for a particular case of indoor 5G scenario, where the presence of an Access Point (AP) was simulated. Coupling the traditional deterministic computational method with an innovative stochastic approach, called Polynomial Chaos Kriging, allowed to evaluate the exposure variability of an user considering the 3D beamforming capability of the antenna. The exposure levels, expressed in terms of specific absorption rate (SAR) in specific tissues, showed low values compared to ICNIRP guidelines.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Pattern recognition
Abstract: In the past decade, the rapid development of machine learning has dramatically improved the performance of epileptic detection with Electroencephalography (EEG). However, only a small amount of labeled epileptic data is available for training because labeling requires numerous neurologists. This paper proposes a one-step semi-supervised epilepsy detection system to reduce the labeling cost by fully utilizing the unlabeled data. The proposed neural network training strategy enables a more robust and accurate decision boundary by forcing the consistency of the double predictions on the same unlabeled data. The results show that the Area Under Receiver Operating Characteristic (AUROC) curves of our proposed model are 10.3% and 4.9% higher than the supervised methods on CHB-MIT and Kaggle datasets, respectively.

Keywords: Causality, Connectivity, Directionality
Abstract: Olfaction and emotions share common networks in the brain. However, little is known on how the emotional content of odors modulate dynamically the cortico-cortical interactions within these networks. In this preliminary study, we investigated the effect of odor valence on effective connectivity through the use of Dynamic Causal Modeling (DCM). We recorded electroencephalographic (EEG) data from healthy subjects performing a passive odor task of odorants with different valence. Once defined a fully-connected a priori network comprising the pyriform cortex (PC), orbitofrontal cortex (OFC), and entorhinal cortex (EC), we tested the modulatory effect of odor valence on their causal interactions at the group level using the parametric empirical bayes (PEB) framework. Results show that both pleasant and the unpleasant odors have an inhibitory effect on the connection from EC to PC, whereas we did not observe any effect for the neutral odor. Moreover, the odor with positive valence has a stronger influence on connectivity dynamics compared to the negative odor. Although preliminary, our results suggest that odor valence can modulate brain connectivity.

Keywords: Physiological systems modeling - Signal processing in physiological systems
Abstract: This study proposes long wave infrared technology as a contactless alternative to wearable devices for stress detection. To this aim, we studied the change in facial thermal distribution of 17 healthy subjects in response to different stressors (Stroop Test, Mental Arithmetic Test). During the experimental sessions the electrodermal activity (EDA) and the facial thermal response were simultaneously recorded from each subject. It is well known from the literature that EDA can be considered a reliable marker for the psychological state variation, therefore we used it as a reference signal to validate the thermal results. Statistical analysis was performed to evaluate significant differences in the thermal features between stress and non-stress conditions, as well as between stress and cognitive load. Our results are in line with the outcomes of previous studies and show significant differences in the temperature trends over time between stress and resting conditions. As a new result, we found that the mean temperature changes of some less studied facial regions, e.g., the right cheek, are able not only to significantly discriminate between resting and stressful conditions, but also allow to recognize the typology of stressors. This outcome not only directs future studies to consider the thermal patterns of less explored facial regions as possible correlates of mental states, but more importantly it suggests that different psychological states could potentially be discriminated in a contactless manner.

Keywords: Neural networks and support vector machines in biosignal processing and classification
Abstract: Abstract— Many biomedical robotic interfaces (e.g., prostheses, exoskeletons) classify or estimate user movement intent based on features extracted from measured electromyograms (EMG). In most cases, the parameters of feature extraction are determined heuristically or assigned arbitrary values. We propose a more rigorous method, numerical optimization, to systematically identify parameters that maximize classification accuracy based on EMG signal characteristics. In this study, we used simulated annealing, a common global numerical optimization method, to find the optimal values of three feature extraction parameters based on the root mean square (rms) magnitude of the EMG signal. The EMG data, obtained from a public database, had been measured from 2 muscles (one hand flexor and one hand extensor) of 5 able-bodied participants performing 6 different movement tasks. Using optimization, we increased the offline movement classification accuracy by 3-5% for each participant and from 79.91% to 92.25% overall. The value of one optimized parameter (threshold of Wilson amplitude) was strongly correlated with the rms magnitude of the EMG signal (R2=0.81). Other parameters were suspected to be related to signal noise, since no strong correlation with rms magnitude was observed. Future studies will refine the optimization approach and test its practicality and effectiveness for improving online classification accuracy with robotic interfaces.

Clinical Relevance— Parameter optimization can potentially make EMG-based control more accurate and reliable by automatically accounting for variations in EMG signal quality across channels or time without changing the data collection procedure.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Time-frequency and time-scale analysis - Wavelets
Abstract: Obstructive sleep apnea (OSA) is a common sleep disorder. Positive airway pressure (PAP) therapy is the first-line treatment, while its effectiveness is significantly limited by incomplete adherence in many patients. This work aims to find a predictive association between data from in-laboratory sleep studies during treatment (PAP titration polysomnogram, or PSG) and PAP adherence. Based on a PAP titration PSG database, we present a pipeline to develop a wavelet-based deep learning model and address two challenges. First, to tackle the problem of extremely long overnight PSG signals, it randomly draws segments and extracts features locally. The global representation for the entire signal is achieved by local feature P-norm pooling. Second, to tackle the problem of limited dataset size, the pre-trained EfficienNet-B7 is used as an unsupervised feature extractor to transfer ImageNet knowledge to PSG signals in the wavelet domain. The trained pipeline achieves 78% balanced accuracy and 83% AUC on the test set using airflow and frontal EEG signals, which, we believe, is a compelling result as a pilot study.

Keywords: Independent component analysis, Principal component analysis
Abstract: In this paper, hand synergies were derived using independent component analysis (ICA) and compared against synergies derived from our previous methods using principal component analysis (PCA). For ICA, we used two algorithms — Infomax and entropy bound minimization (EBM). For all the methods, the synergies were extracted from rapid hand grasps. The extracted synergies were then tested for generalizability in reconstructing natural hand grasps and American Sign Language (ASL) postures that were different from rapid grasps. The results indicate that the synergies derived from ICA were able to generalize only marginally better when compared to those from PCA. Among the two ICA methods, Infomax performed slightly better in yielding lower reconstruction error while EBM performed better in sparse selection of synergies. The implications and future scope were discussed.

Keywords: Data mining and big data methods - Biosignal classification, Data mining and big data methods - Machine learning and deep learning methods, Physiological systems modeling - Signal processing in physiological systems
Abstract: The voltage criteria used to diagnose left ventricular hypertrophy (LVH) in the chest and limb leads are by no means absolute. In addition to QRS voltages, QRS axis and duration, and P wave characteristics, repolarization (ST-T) changes have been focused attention due to their representing left ventricular overload. Vectorcardiography (VCG) has been studied specifically on its repolarization abnormality. The present study aims to devise spatial feature extraction of VCG and assess it in the LVH classification task. A minimum volume ellipsoid enclosure was applied to six segments obtained from upstroke and downstroke of each P, QRS, and T loop of a single-beat VCG. For the evaluation, VCG and 12 lead ECG datasets along with LVH labels of 61 subjects were derived from public open data, PTB-XL. These classification performances were compared with the LVH diagnosis criteria in the standard 12 lead ECG. As a result, the Random Forest classifier trained by the proposed spatial VCG feature resulted in an accuracy of 0.904 (95% confidence interval: 0.861-0.947) when the class-balanced dataset was evaluated, which slightly exceeded the feature of 12 lead ECG. The feature importance analysis provided the quantitative ranking of the spatial feature of VCG, which was practically similar to those of ECG in the LVH classification task. Since the VCG are spatially comparable with three-dimensional data of CT, MRI, or Echocardiography, VCG will shed light on the spatial behavior of electrical depolarization and repolarization abnormalities in cardiac diseases.

Keywords: Data mining and big data methods - Biosignal classification, Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Several studies have demonstrated that error-related neuronal signatures can be successfully detected and used to improve the performance of brain-computer interfaces. However, this has been tested mainly in well-controlled environments and based on temporal features, such as the amplitude of event-related potentials. In this study, we propose a classification algorithm combining frequency features and a weighted SVM to detect the neuronal signatures of errors committed in a complex saccadic go/no-go task. We follow the hypothesis that frequency features yield better discrimination performance in complex tasks, generalize better, and require fewer pre-processing steps. When combining temporal and frequency features, we achieved a balanced classification accuracy of 75% - almost the same as using only frequency features. On the other hand, when using only temporal features, the balanced accuracy decreased to 66%. These findings show that subjects’ performance can be automatically detected based on frequency features of error-related neuronal signatures. Additionally, our results revealed that features computed in the pre-response time contribute to the discrimination between correct and erroneous responses, which suggests the existence of error-related patterns even before response execution.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification
Abstract: We propose a divide-and-conquer approach to detect depression severity using speech. We divide speech features based on their attributes, i.e., acoustic, prosodic, and language features, then fuse them in a modeling stage with fully connected deep neural networks. Experiments with 76 clinically depressed patients (38 severe and 38 moderate in terms of Montgomery-Asberg Depression Rating Scale (MADRS)), we obtain 78% accuracy while patients’ self-reporting scores can classify their own status with 79% accuracy.

Keywords: Signal pattern classification, Time-frequency and time-scale analysis - Nonstationary analysis and modeling, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Surface electromyography (sEMG) signals are now commonly used in continuous myoelectric control of prostheses. More recently, researchers have considered EMG-based gesture recognition systems for human computer interaction research. These systems instead focus on recognizing discrete gestures (like a finger snap). The majority of works, however, have focused on improving multi-class performance, with little consideration for false activations from "other" classes. Consequently, they lack the robustness needed for real-world applications which generally require a single motion class such as a mouse click or a wake word. Furthermore, many works have borrowed the windowed classification schemes from continuous control, and thus fail to leverage the temporal structure of the gesture. In this paper, we propose a verification-based approach to creating a robust EMG wake word using one-class classifiers (Support Vector Data Description, One Class-Support Vector Machine, Dynamic Time Warping (DTW) & Hidden Markov Models). The area under the ROC curve (AUC) is used as a feature optimization objective as it provides a better representation of the verification performance. Equal error rate (EER) and AUC are then used as evaluation metrics. The results are computed using both window-based and temporal classifiers on a dataset consisting of five different gestures, with a best EER of 0.04 and AUC of 0.98, recorded using a DTW scheme. These results demonstrate a design framework that may benefit the development of more robust solutions for EMG-based wake words or input commands for a variety of interactive applications.

Keywords: Signal pattern classification, Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Biosignal classification
Abstract: Patient independent epileptic seizure detection algorithm for scalp electroencephalogram (EEG) data is proposed in this paper. Principal motivation of this work is to integrate neural and conventional machine learning methods to develop a classification system which can advance the current wearable health systems in terms of computational complexity and accuracy. Being based on processing a single channel EEG processing, the approach is suitable for usage with small wireless sensors. A shallow autoencoder model is utilized for sparse representation of the EEG signal followed by k-nearest neighbor (kNN) classifier to categorize the data as epileptic or non-epileptic. Using a single EEG channel an optimum sparsity level is explored in the encoded sample. Attaining an accuracy, sensitivity and specificity of 98.85%, 99.29% and 98.86% respectively, for CHB-MIT scalp EEG database, proposed classification method outperforms state of the art seizure detection methodologies. Experiments has shown that this performance was possible by using a sparsity level of 4 in the auto encoder. Furthermore, use of shallow learning instead of deep learning approach for generation of sparse but effective representation is computationally lighter than many other feature extraction and preprocessing methods.

Keywords: Time-frequency and time-scale analysis - Nonstationary analysis and modeling, Physiological systems modeling - Signal processing in physiological systems
Abstract: Heart rate variability (HRV) is a non-stationary, irregularly sampled signal that represents changes in heart rate over time. The HRV spectrum can be divided into four main ranges covering high, low, very low and ultra-low frequencies. The components lying in these bands, both amplitude and frequency modulated, provide valuable information about various physiological processes. The aim of this study was to verify the usefulness of adaptive variational mode decomposition (AVMD) in the extraction of these components from overnight HRV. The effectiveness of this new approach was compared to multiband filtering (MBF) based on a synthetically generated signal, as well as real data from three patients. AVMD turned out to be more robust and effective than MBF, particularly in the high and low frequency ranges, making it a reliable method for deriving the HRV frequency components.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Depression is a common and serious mental illness which negatively affects daily functioning. To prevent the progression of the illness into severe or long-term consequences, early diagnosis is crucial. We developed an automated speech feature analysis application for depression and other psychiatric disorders derived from a developed Thai psychiatric and verbal screening test. The screening test includes Thai's version of Patient Health Questionnaire-9 (PHQ-9) and Hamilton Depression Rating Scale (HAM-D), and 32 additional emotion-induced questions. Case-control study was conducted on speech features from 66 participants. Twenty seven of those had depression (DP), 12 had other psychiatric disorders (OP), and 27 were normal controls (NC). The five-fold cross-validation from 6 settings of 5 classifiers with the combination of PHQ-9 and HAM-D scores, and speech features were examined. Results showed highest performance from the multilayer perceptron (MLP) classifier which yielded 83.33% sensitivity, 91.67% specificity, and 83.33% accuracy, where negative-emotional questions were most effective in classification. The automated speech feature analysis showed promising results for screening patients with depression or other psychiatric disorders. The current application is accessible through smartphone, making it a feasible and intuitive setup for low-resource countries such as Thailand.

Keywords: Multivariate methods, Adaptive filtering
Abstract: In this paper, we propose to learn a spatial filter directly from ElectroEncephaloGraphy (EEG) signals using Graph Signal Processing tools. We combine a graph learning algorithm with a high-pass graph filter to remove spatially large signals from the raw data. This approach increases topographical localization, and attenuates volume-conducted features. We empirically show that our method gives similar results that the surface Laplacian in the noiseless case while being more robust to noise or defective electrodes.

Clinical relevance — The proposed method is an alternative to the surface Laplacian filter that is commonly used for processing EEG signals. It could be used in cases where this standard approach does not provide satisfying results (low signal-to-noise ratios due to a low number of epochs, defective electrodes). This could be particularly interesting in case of an electrode defect, as it can happen in clinical practice.

Keywords: Time-frequency and time-scale analysis - Wavelets, Data mining and big data methods - Machine learning and deep learning methods, Physiological systems modeling - Signal processing in physiological systems
Abstract: Analysis and classification of electromyography (EMG) signals are crucial for rehabilitation and motor control. This paper investigates electromyogram (EMG) time-frequency representations and then creates conventional and deep learning models for EMG signal classification. Firstly, a dataset of single-channel surface EMG signals has been recorded for four subjects to differentiate between forearm flexion and extension. Then, different time-frequency EMG representations have been used to build conventional and deep learning models for EMG classification. We compared the performance of pre-trained convolutional neural network models, namely GoogLeNet, SqueezeNet and AlexNet, and achieved accuracies of 92.71%, 90.63% and 87.5%, respectively. Also, data augmentation techniques on the levels of raw EMG signals and their time-frequency representations helped improve the accuracy of GoogLeNet to 96.88%. Furthermore, our approach demonstrated superior performance on another publicly available 10-class EMG dataset, and also using traditional classifiers trained on hand-crafted features.

Keywords: Time-frequency and time-scale analysis - Nonstationary analysis and modeling, Physiological systems modeling - Multivariate signal processing, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Accurate torque estimation during dynamic conditions is challenging, yet an important problem for many applications such as robotics, prosthesis control, and clinical diagnostics. Our objective is to accurately estimate the torque generated at the elbow during flexion and extension, under quasi-dynamic and dynamic conditions. High-density surface electromyogram (HD-EMG) signals, acquired from the long head and short head of biceps brachii, brachioradialis, and triceps brachii of five participants are used to estimate the torque generated at the elbow, using a convolutional neural network (CNN). We hypothesise that incorporating the mechanical information recorded by the biodex machine, i.e., position and velocity, can improve the model performance. To investigate the effects of the added data modalities on the model accuracy, models are constructed that combine EMG and position, as well as EMG with both position and velocity. R2 values are improved by 2.35%, 37.50%, and 16.67%, when position and EMG are used as inputs to the CNN models, for isotonic, isokinetic, and dynamic cases, respectively compared to using only EMG. The model performances improves further by 2.29%, 12.12%, and 20.50% for isotonic, isokinetic, and dynamic conditions, when velocity is added with the EMG and position data.

Keywords: Neural networks and support vector machines in biosignal processing and classification
Abstract: Cardiovascular (CV) diseases are the leading cause of death in the world, and auscultation is typically an essential part of a cardiovascular examination. The ability to diagnose a patient based on their heart sounds is a rather difficult skill to master. Thus, many approaches for automated heart auscultation have been explored. However, most of the previously proposed methods involve a segmentation step, the performance of which drops significantly for high pulse rates or noisy signals. In this work, we propose a novel segmentation-free heart sound classification method. Specifically, we apply discrete wavelet transform to denoise the signal, followed by feature extraction and feature reduction. Then, Support Vector Machines and Deep Neural Networks are utilised for classification. On the PASCAL heart sound dataset our approach showed superior performance compared to others, achieving 81% and 96% precision on normal and murmur classes, respectively. In addition, for the first time, the data were further explored under a user-independent setting, where the proposed method achieved 92% and 86% precision on normal and murmur, demonstrating the potential of enabling automatic murmur detection for practical use.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Physiological systems modeling - Signal processing in simulation, Nonlinear dynamic analysis - Biomedical signals
Abstract: High-density surface electromyography (EMG) has been proposed to overcome the lower selectivity with respect to needle EMG and to provide information on a wide area over the considered muscle. Motor units decomposed from surface EMG signal of different depths differ in the distribution of action potentials detected in the skin surface. We propose a noninvasive model for estimating the depth of motor unit. We find that the depth of motor unit is linearly related to the Gaussian RMS width fitted by data points extracted from motor unit action potential. Simulated and experimental signals are used to evaluate the model performance. The correlation coefficient between reference depth and estimated depth is 0.92 ± 0.01 for simulated motor unit action potentials. Due to the symmetric nature of our model, no significant decrease is detected during the electrode selection procedure. We further checked the estimation results from decomposed motor units, the correlation coefficient between reference depth and estimated depth is 0.82 ± 0.07. For experimental signals, high discrimination of estimated depth vector is detected across gestures among trials. These results show the potential for a straightforward assessment of depth of motor units inside muscles. We discuss the potential of a non-invasive way for the location of decomposed motor units.

Keywords: Physiological systems modeling - Closed loop systems
Abstract: Post-prandial hypoglycemia occurs 2-5 hours after food intake, in not only insulin-treated patients with diabetes but also other metabolic disorders. For example, postprandial hypoglycemia is an increasingly recognized late metabolic complication of bariatric surgery (also known as PBH), particularly gastric bypass. Underlying mechanisms remain incompletely understood to date. Besides excessive insulin exposure, impaired counter-regulation may be a further pathophysiological feature. To test this hypothesis, we need standardized postprandial hypoglycemic clamp procedures in affected and unaffected individuals allowing to reach identical predefined postprandial hypoglycemic trajectories. Generally, in these experiments, clinical investigators manually adjust glucose infusion rate (GIR) to clamp blood glucose (BG) to a target hypoglycemic value. Nevertheless, reaching the desired target by manual adjustment may be challenging and possible glycemic undershoots when approaching hypoglycemia can be a safety concern for patients. In this study, we developed a PID algorithm to assist clinical investigators in adjusting GIR to reach the predefined trajectory and hypoglycemic target. The algorithm is developed in a manual mode to permit the clinical investigator to interfere. We test the controller in silico by simulating glucose-insulin dynamics in PBH and healthy non-surgical individuals. Different scenarios are designed to test the robustness of the algorithm to different sources of variability and to errors, e.g. outliers in the BG measurements, sampling delays or missed measurements. The results prove that the PID algorithm is capable to accurately and safely reaching the target BG level, on both healthy and PBH subjects, with a median deviation from the reference of 2.8% and 2.4% respectively.

Keywords: Parametric filtering and estimation
Abstract: Existing literature suggests that the probability density function (pdf) of surface Electromyography (sEMG) signals follows either a Gaussian or Laplacian model. In this paper, a Laplacian-Gaussian mixture model is proposed for the EMG signals extracted from the upper limbs. The model is validated using both quantitative and qualitative perspectives. Specifically, for a benchmark dataset, the Kullback–Leibler (KL) divergence is computed between the proposed model and the histogram based empirical probability density function (mpdf). For a sample signal, a goodness of fit plot with R squared value and a visual comparison between the histogram based mpdf and the estimated pdf from the proposed model are presented. Moreover, the Expectation-Maximization (EM) algorithm is derived for the estimation of the parameters of the proposed mixture model. The weight of the Laplacian component is computed for each of the signals from a benchmark dataset. It has been empirically determined that the Laplacian component has a major contribution to the mixture.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: The automated recognition of human emotions plays an important role in developing machines with emotional intelligence. However, most of the affective computing models are based on images, audio, videos and brain signals. There is a lack of prior studies that focus on utilizing only peripheral physiological signals for emotion recognition, which can ideally be implemented in daily life settings using wearables, e.g., smartwatches. Here, an emotion classification method using peripheral physiological signals, obtained by wearable devices that enable continuous monitoring of emotional states, is presented. A Long Short-Term Memory neural network-based classification model is proposed to accurately predict emotions in real-time into binary levels and quadrants of the arousal-valence space. The peripheral sensored data used here were collected from 20 participants, who engaged in a naturalistic debate. Different annotation schemes were adopted and their impact on the classification performance was explored. Evaluation results demonstrate the capability of our method with a measured accuracy of >93% and >89% for binary levels and quad classes, respectively. This paves the way for enhancing the role of wearable devices in emotional state recognition in everyday life.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Thammasat-NECTEC-Chula’s Thai Language and Cognition Assessment (TLCA) is a cognitive paper-based test consisting of 21 tasks that cover 3 domains: memory, language, and other cognitive abilities. The TLCA follows some aspects of the existing tests (Thai Addenbrooke’s Cognitive Examination-Revised (Thai-ACE-R) and the Thai Montreal Cognitive Assessment Test (Thai-MoCA)) and many parts were reconstructed to be more adapted to the Thai culture. Data obtained from the test will be able to precisely distinguish between patients with Mild Cognitive Impairment (MCI), Alzheimer’s Disease (AD), and Normal healthy Controls (NC). The TLCA was tested on 90 participants (32 on the paper-based version and 58 on the computerized version) using a scoring procedure and speech features from verbal responses with machine learning classification. The scoring results showed significant difference between non-AD (NC + MCI) vs AD participants in 3 domains and could differentiate between NC and MCI, while machine classification could classify in three settings: NC vs non-NC (MCI + AD), AD vs non-AD and NC vs MCI vs AD. These promising results suggest that TLCA could be further verified and used as an efficient assessment in MCI and AD screening for Thais.

Keywords: Signal pattern classification, Parametric filtering and estimation, Time-frequency and time-scale analysis - Wavelets
Abstract: In this work, we propose an unsupervised algorithm for fundamental heart sound detection. We propose to detect the heart sound candidates using the stationary wavelet transforms and group delay. We further propose an objective function to select the candidates. The objective function has two parts. We model the energy contour of S1/S2 sound using the Gaussian mixture function (GMF). The goodness of fit for the GMF is used as the first part of the objective function. The second part of the objective function captures the consistency of the heart sounds’ relative location. We solve the objective function efficiently using dynamic programming. We evaluate the algorithm on Michigan HeartSound and Murmur database. We also assess the algorithm’s performance using the three different additive noises– white Gaussian noise (AWGN), Student-t noise, and impulsive noise. The experiments demonstrate that the proposed method performs better than baseline in both clean and noisy conditions. We found that the proposed method is robust in the case of AWGN noise and student-t distribution noise. But its performance reduces in the case of impulsive noise.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification
Abstract: Continuous kinematics estimation from surface electromyography (sEMG) allows more natural and intuitive human-machine collaboration. Recent research has suggested the use of multimodal inputs (sEMG signals and inertial measurements) to improve estimation performance. This work focused on assessing the use of angular velocity in combination with myoelectric signals to simultaneously and continuously predict 12 joint angles in the hand. Estimation performance was evaluated for five functional and grasping movements in 20 subjects. The proposed method is based on convolutional and recurrent neural networks using transfer learning (TL). A novel aspect was the use of a pretrained deep network model from basic joint hand movements to learn new patterns present in functional motions. A comparison was carried out with the traditional method based solely on sEMG. Although the performance of the algorithm slightly improved with the use of the multimodal combination, both strategies had similar behavior. The results indicated a significant improvement for a single task: opening a bottle with a tripod grasp.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Physiological systems modeling - Multivariate signal processing
Abstract: Obstructive Sleep Apnea (OSA) is a sleep disorder associated with reduced vigilance. Vigilance status is often measured using the Psychomotor Vigilance Task (PVT). This paper investigates modelling strategies to map sleep spindle (Sp) characteristics to PVT metrics in patients with OSA. Sleep spindles (n=2305) were manually detected across blocks of sleep for 20 patients randomly selected from a cohort of 190 undergoing Polysomnography (PSG) for suspected OSA. Novel Sp metrics based on runs or “bursts” of Sps were used to model Sp characteristics to standardized (z) Lapse and Median Reaction Time (MdRT) scores, and to Groups based on zLapse and zMdRT scores. A model employing Sp Burst characteristics mapped to zMdRT Group membership with an accuracy of 91.9%, (95% C.I. 90.8-93.0). The model had a sensitivity of 88.9%, (95% C.I. 87.5-89.0) and specificity of 89.1% (95% C.I. 87.3-90.5) for detecting patients with the lowest MdRTs in our cohort.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Nonlinear dynamic analysis - Biomedical signals
Abstract: The quality of the extracted traditional hand-crafted Electromyogram (EMG) features has been recently identified in the literature as a limiting factor prohibiting the translation from laboratory to clinical settings. To address this limitation, a shift of focus from traditional feature extraction methods to deep learning models was witnessed, as the latter can learn the best feature representation for the task at hand. However, while deep learning models achieve promising results based on raw EMG data, their clinical implementation is often challenged due to their significantly high computational costs (significantly large number of generated models’ parameters and a huge amount of data needed for training). This paper aims to focus on combining the simplicity and low computational characteristics of traditional feature extraction with the memory concepts from Long Short-Term Memory (LSTM) models to efficiently extract the spatial-temporal dynamics of the EMG signals. The novelty of the proposed method can be summarized in a) the memory concept leveraged from deep learning structures, capturing short-term temporal dependencies of the EMG signals, b) the use of cardinality to generate logical combinations of spatially distinct EMG signals and as a feature extraction method and 3) low computational costs and the enhanced classification performance. The performance of the proposed method is validated using three EMG datasets from laboratory hardware: 9 transradial amputees and 17 intact-limbed, and wearables: 22 intact-limed using two wearable consumer armbands. The proposed method shows significantly myoelectric pattern recognition performance, with accuracies reaching up to 99%.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Biosignal classification
Abstract: Cardiac Auscultation, an integral part of the physical examination of a patient, is essential for early diagnosis of cardiovascular diseases (CVDs). The ability to accurately diagnose the heart sounds requires experience and expertise, which is lacking in doctors in the early years of clinical practice. Thus, there is a need for an automatic diagnostic tool that would aid medical practitioners with their diagnosis. We propose novel hybrid architectures for classification of unsegmented heart sounds to normal and abnormal classes. We propose two methods, with and without the conventional feature extraction step in the classification pipeline. We demonstrate that the F score using the approach with conventional feature extraction is 1.25 (absolute) more than using a baseline implementation on the Physionet dataset. We also introduce a mechanism to tag predictions as unsure and compare results with a varying threshold.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Signal pattern classification
Abstract: Low-cost wearables with capability to record electrocardiograms (ECG) are becoming increasingly available. These wearables typically acquire single-lead ECGs that are mainly used for screening of cardiac arrhythmias such as atrial fibrillation. Existing R-peak detection methods are fairly accurate but have limited precision. To enable data-point precise detection of R-peaks, we propose a method that uses a fully convolutional dilated neural network. The network is trained and evaluated with manually annotated R-peaks in a heterogeneous set of ECGs that contain a wide range of cardiac rhythms and acquisition noise. 700 randomly chosen ECGs from the PhysioNet/CinC challenge 2017 were used for training (n=500), validation (n=100) and testing (n=100). The network achieves a precision of 0.910, recall of 0.926, and an F1-score of 0.918 on the test set. Our data-point precise R-peak detector may be important step towards fully automatic cardiac arrhythmia detection.

Keywords: Connectivity
Abstract: Connectivity analyses are widely used to assess the interaction brain networks. This type of analyses is usually conducted considering the well-known classical frequency bands: delta, theta, alpha, beta, and gamma. However, this parcellation of the frequency content can bias the analyses, since it does not consider the between-subject variability or the particular idiosyncrasies of the connectivity patterns that occur within a band. In this study, we addressed these limitations by introducing the High Frequential Resolution Networks (HFRNs). HFRNs were constructed, using a narrow-bandwidth FIR bank filter of 1 Hz bandwidth, for two different connectivity metrics (Amplitude Envelope Correlation, AEC, and Phase Lag index, PLI) and for 3 different databases of MEG and EEG recordings. Results showed a noticeable similarity between the frequential evolution of PLI, AEC, and the Power Spectral Density (PSD) from MEG and EEG signals. Nonetheless, some technical remarks should be considered: (i) results at the gamma band should exclude the frequency range around 50 Hz due to abnormal connectivity patterns, consequence of the previously applied 50 Hz notch-filter; (ii) HFRNs patterns barely vary with the connection distance; and (iii) a low sampling frequency can exert a remarkable influence on HFRNs. To conclude, we proposed a new framework to perform connectivity analyses that allow to further analyze the frequency-based distribution of brain networks.

Keywords: Data mining and big data methods - Biosignal classification, Data mining and big data methods - Pattern recognition, Signal pattern classification
Abstract: Phantom Limb Pain (PLP) is a chronic condition frequent among individuals with acquired amputation. PLP has been often investigated with the use of functional MRI focusing on the changes that take place in the sensorimotor cortex after amputation. In the present study, we investigated whether a different type of data, namely electroencephalographic (EEG) recordings, can be used to study the condition. We acquired resting state EEG data from people with and without PLP and then used machine learning for a binary classification task that differentiates the two. Common Spatial Pattern decomposition was used as the feature extraction method and two validation schemes were followed for the classification task. Six classifiers (LDA, Log, QDA, LinearSVC, SVC and RF) were optimized through grid search and their performance compared. Two validation approaches, namely all-subjects validation and leave-one-out cross-validation (LOOCV), resulted in high classification accuracy. Most notably, the 93.7% accuracy achieved with SVC in LOOCV holds promise for good diagnostic capabilities using EEG biomarkers. In conclusion, our findings indicate that EEG data is a promising target for future research aiming at elucidating the neural mechanisms underlying PLP and its diagnosis.

Keywords: Physiological systems modeling - Signal processing in simulation, Connectivity, Directionality
Abstract: Catheter ablation for atrial fibrillation (AF) is one of the most commonly performed electrophysiology procedures. Despite significant advances in our understanding of AF mechanisms in the last years, ablation outcomes remain suboptimal for many patients, particularly those with persistent or long-standing AF. A possible reason is that ablation techniques mainly focus on anatomic, rather than patient-specific functional targets for ablation. The identification of such ablation targets remains challenging. The purpose of this study is to investigate a novel approach based on directed networks, which allow the automatic detection of important arrhythmia mechanisms, that can be convenient for guiding the ablation strategy. The networks are generated by processing unipolar electrograms (EGMs) collected by the catheters positioned at the different regions of the atria. Network vertices represent the locations of the recordings and edges are determined using cross-covariance time-delay estimation method. The algorithm identifies rotational activity, spreading from vertex to vertex creating a cycle. This work is a simulation study and it uses a highly detailed computational 3D model of human atria in which sustained rotor activation of the atria was achieved. Virtual electrodes were placed on the endocardial surface, and EGMs were calculated at each of these electrodes. The propagation of the electric wave fronts in the atrial myocardium during AF is very complex, so in order to properly capture wave propagation patterns, we split EGMs into multiple short time frames. Then, a specific network for each of these time frames was generated, and the cycles repeating in consecutive networks point us to the stable rotor's location. The respective atrial voltage map served as reference. By detecting a cycle between the same 3 nodes in 19 out of 58 networks, where 10 of these networks were in consecutive time frames, a stable rotor was successfully located.

Keywords: Signal pattern classification, Time-frequency and time-scale analysis - Time-frequency analysis, Principal component analysis
Abstract: Cardiovascular diseases(CVDs) are the world’s leading cause of death. Endothelial Dysfunction is an early stage of cardiovascular diseases and can effectively be used to detect the presence of the CVDs, monitor its progress and investigate the effectiveness of the treatment given. This study proposes a reliable approach for the screening of endothelial dysfunction via machine learning, using features extracted from a combination of Plethysmography, Digital Thermal Monitoring, biological features (age and gender) and anthropometry (BMI and pulse pressure). This case control study includes 55 healthy subjects and 45 subjects with clinically verified CVDs. Following the feature engineering stage, the results were subjected to dimension reduction and 5-fold cross-validation where it was observed that models Logistic Regression and Linear Discriminant provided the highest accuracies of 84% and 81% respectively. We propose that this study can be used as an efficient guide for the non-invasive screening of endothelial dysfunction

Keywords: Physiological systems modeling - Signal processing in physiological systems
Abstract: Food variety influences appetitive behaviour, motivation to eat and energy intake. Research found that repeated exposure to varied food images increases the motivation towards food in adults and children. This study investigates the effects of repetition on the modulation of early and late components of event-related potentials (ERPs) when participants passively viewed the same food and non-food images repeatedly. The motivational attention to food and non-food images were assessed in frontal, centroparietal, parietooccipital and occipitotemporal areas of the brain. Participants showed increased late positive potential (late ERP component) to high caloric image in the occipitotemporal region compared to low caloric and nonfood images. Similar effects could be seen in the early ERP component in the frontal region, but with reversed polarity. Data suggest that both the early and late ERP components show greater ERP amplitude when viewing high caloric images more than low caloric and non-food images. Despite repeated exposure to same image, high caloric food continued to show sustained attention compared to low caloric and non-food image.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Respiratory rate (RR) is a clinical sign representing ventilation. An abnormal change in RR is often the first sign of health deterioration as the body attempts to maintain oxygen delivery to its tissues. There has been a growing interest in remotely monitoring of RR in everyday settings which has made photoplethysmography (PPG) monitoring wearable devices an attractive choice. PPG signals are useful sources for RR extraction due to the presence of respiration-induced modulations in them. The existing PPG-based RR estimation methods mainly rely on hand-crafted rules and manual parameters tuning. An end-to-end deep learning approach was recently proposed, however, despite its automatic nature, the performance of this method is not ideal using the real world data. In this paper, we present an end-to-end and accurate pipeline for RR estimation using Cycle Generative Adversarial Networks (CycleGAN) to reconstruct respiratory signals from raw PPG signals. Our results demonstrate a higher RR estimation accuracy of up to 2times (mean absolute error of 1.9pm0.3 using five fold cross validation) compared to the state-of-th-art using a identical publicly available dataset. Our results suggest that CycleGAN can be a valuable method for RR estimation from raw PPG signals.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Directionality, Multivariate methods
Abstract: Heart Period (H) results from the activity of several coexisting control mechanisms, involving Systolic Arterial Pressure (S) and Respiration (R), which operate across multipletime scales encompassing not only short term dynamics but also long-range correlations. In this work, multiscale representation of Transfer Entropy (TE) and of its decomposition in the network of these three interacting processes is obtained by extending the multivariate approach based on linear parametric VAR models to the Vector AutoRegressive Fractionally Integrated (VARFI) framework for Gaussian processes. This approach allows to dissect the different contributions to cardiac dynamics accounting for the simultaneous presence of short and long term dynamics. The proposed method is first tested on simulations of a benchmark VARFI model and then applied to experimental data consisting of H, S and R time series measured in healthy subjects monitored at rest and during mental and postural stress. The results reveal that the proposed method can highlight the dependence of the information transfer on the balance between short-term and long-range correlations in coupled dynamical systems.

Keywords: Signal pattern classification, Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis
Abstract: Mechanical ventilation is necessary to maintain patients’ life in intensive care units. However, too early or too late extubation may injure the muscles or lead to respiratory failure. Therefore, the spontaneous breathing trial (SBT) is applied for testing whether the patients can spontaneously breathe or not. However, previous evidence still reported 15%~20% of the rate of extubation fail. The monitor only considers the ventilation variables during SBT. Therefore, this study measures the asynchronization between thoracic and abdomen wall movement (TWM and AWM) by using instantaneous phase difference method (IPD) during SBT for 120 minutes. The respiratory inductive plethysmography were used for TWM and AWM measurement. The preliminary result recruited 31 signals for further analysis. The result showed that in successful extubation group can be classified into two groups, IPD increase group, and IPD decrease group; but in extubation fail group, the IPD value only increase. Therefore, the IPD decrease group can almost perfectly be discriminated with extubation fail group, especially after 70 minutes (Area under curve of operating characteristic curve was 1). These results showed IPD is an important key factor to find whether the patient is suitable for extubation or not. These finding suggest that the asynchronization between TWM and AWM should be considered as a predictor of extubation outcome. In future work, we plan to recruit 150 subjects to validate the result of this preliminary result. In addition, advanced machine learning method is considered to apply for building effective models to discriminate the IPD increase group and extubation fail group.

Keywords: Kalman filtering, Physiological systems modeling - Signal processing in physiological systems, Physiological systems modeling - Closed loop systems
Abstract: Stress has effects on productivity and performance. Poor stress management may lead to reduced productivity and performance. Non-invasive actuators such as music have the potential to effectively regulate stress. In this study, using a state-space approach, we obtain a performance state to investigate the performance during a working memory task while playing two different types of music in the background. In our experiments, participants performed a working memory task while listening to calming and vexing music of their choice. We utilize the binary correct/incorrect response and the continuous reaction time of the response from the participants to quantify the performance. The state-space quantification reveals that vexing music has a statistically significant positive impact on the obtained performance state. This indicates the feasibility of designing non-invasive closed-loop systems to regulate stress for maximizing performance and productivity.