BIOSIGNALS 2021 Abstracts


Full Papers
Paper Nr: 2
Title:

Evaluation of Stair Climbing as an Approach for Estimating Heart Rate Recovery in Daily Activities

Authors:

Daivaras Sokas, Andrius Rapalis, Andrius Petrėnas, Saulius Daukantas and Vaidotas Marozas

Abstract: Post-Exercise heart rate recovery (HRR) is a convenient approach to assess cardiovascular autonomic function. Ordinary stair climbing can be viewed as an alternative HRR test performed in daily activities, and also well-suited for implementation in wrist-worn devices. This study compares HHR parameters estimated after stair climbing to those obtained by performing the conventional YMCA bench step test using a custom-made wrist-worn device and a consumer smart wristband Fitbit Charge 2. The results show that most HHR parameters are underestimated after stair climbing but still comparable to those obtained from the bench step test. The lowest relative error, 8–11% on average, was found for the decay of heart rate in 30, 60, and 120 s after the recovery onset.

Paper Nr: 9
Title:

Detection of Error Correlates in the Motor Cortex in a Long Term Clinical Trial of ECoG based Brain Computer Interface

Authors:

Vincent Rouanne, Maciej Śliwowski, Thomas Costecalde, Alim L. Benabid and Tetiana Aksenova

Abstract: Error correlates are thought to be promising for BCIs as a way to perform error correction or prevention, or to label data in order to perform online adaptation of BCIs’ control models. Current state-of-the-art BCIs are motor-imagery-based invasive BCIs and thus have no access to neural data apart from sensory-motor cortices. We investigated at the single trial level the presence and detectability of error correlates in the primary motor cortex during observation or motor imagery (MI) control of a BCI with two discrete classes by a tetraplegic user. We show that error correlates can be detected using a broad range of classifiers, namely Support Vector Machine (SVM), logistic regression, N-way Partial Least Squares (NPLS), Multilayer Perceptron (MLP) and Convolutional Neural Network (CNN) with respective mean AUC of the ROC curve of 0.645, 0.662, 0.642, 0.680 and 0.630 in the observation condition, and 0.623, 0.605, 0.603, 0.626 and 0.580 in the MI-control condition. We also suggest that these error correlates are stable in time. These findings suggest that error correlates could be used in clinical trials using invasive motor-imagery-based BCIs for error correction or prevention.

Paper Nr: 10
Title:

Chaotic Changes in Fingertip Pulse Waves during Autobiographical Memory Retrieval

Authors:

Eri Shibayama and Taira Suzuki

Abstract: The effects of autobiographical memory retrieval on a psychophysiological index were examined. The experimental (retrieval) group conducted an autobiographical memory retrieval task, whereas the control group repeatedly vocalized Japanese syllabary. The largest fingertip pulse-wave Lyapunov Exponent (LLE) in chaos analysis, which is a nonlinear analysis, was measured as an objective psychophysiological index. Moreover, participants responded to a psychological questionnaire and completed an original checklist before and after the experiment. The results indicated that fingertip pulse-wave LLE increased significantly only in the retrieval group. This result supported previous findings that psychotherapies such as reminiscence therapy have positive effects on cognitive functions at the psychophysiological level. Moreover, only the retrieval group showed significant improvements in psychological indices. Therefore, recalling autobiographical memories and verbally sharing it with others might contribute to maintaining mental health. To date, autobiographical memory retrieval and reminiscence therapy have not been sufficiently examined by using physiological indices. This study’s results using a physiological index are expected to contribute to research on reminiscence therapy.

Paper Nr: 11
Title:

Elliptical Fitting as an Alternative Approach to Complex Nonlinear Least Squares Regression for Modeling Electrochemical Impedance Spectroscopy

Authors:

Norman Pfeiffer, Toni Wachter, Jürgen Frickel, Christian Hofmann, Abdelhamid Errachid and Albert Heuberger

Abstract: Electrochemical impedance spectroscopy is an important procedure with the ability to describe a wide range of physical and chemical properties of electrochemical systems. The spectral behavior of impedimetric sensors is mostly described by the Randles circuit, whose parameters are determined by regression techniques on the basis of measured spectra. The charge transfer resistance as one of these parameters is often used as sensor response. In the laboratory environment, the regression is usually performed by commercial software, but for integrated, application-oriented solutions, separate approaches must be pursued. This work presents an approach for elliptical fitting of the curve in the Nyquist plot, which is compared to the complex nonlinear least squares (CNLS) regression technique. For this purpose, artificial spectra were generated, which were considered both with and without noise superposition. Although the average error in calculating the charge transfer resistance from noisy signals using the elliptical fitting of −2.7% was worse than the CNLS with 2.4 · 10−2%, the former required only about 1/225 of the computing time compared to the latter. Following application-oriented evaluations of the achievable accuracies, the elliptical approach may turn out to be a resource saving alternative.

Paper Nr: 12
Title:

Choosing the Appropriate QRS Detector

Authors:

Justus Eilers, Jonas Chromik and Bert Arnrich

Abstract: QRS detectors are used as the most basic processing tool for ECG signals. Thus, there are many situations and signals with a wide range of characteristics in which they shall show great performance. Despite the expected versatility, most of the published QRS detectors are not tested on a diverse dataset. Using 14 databases, 10,000 heartbeats for each different heartbeat type were extracted to show that there are notable performance differences for the tested eight algorithms. Besides the analysis on heartbeat types, this paper also tests the noise resilience regarding different noise combinations. Each of the tested QRS detectors showed significant differences depending on heartbeat type and noise combination. This leads to the conclusion that before choosing a QRS detector, one should consider its use case and test the detector on data representing it. For authors of QRS detectors, this means that every algorithm evaluation should employ a dataset that is as diverse as the one used in this paper to assess the QRS detector’s performance in an objective and unbiased manner.

Paper Nr: 14
Title:

Characterization of Upper Limb Functionality Caused by Neuromuscular Disorders using Novel Motion Features from a Specialized Gaming Platform

Authors:

A. Chytas, D. Fotopoulos, V. Kilintzis, E. Koutsiana, I. Ladakis, E. Kiana, T. Loizidis and I. Chouvarda

Abstract: This paper describes the methodology for analyzing upper limb motion data derived from a novel Gamified Motion Control Assessment platform that is based on a virtual 3D game environment. The gamified approach targets patients experiencing upper-limb movement hindrances, typically caused by neuromuscular disorders. The leap motion controller is used for interaction. The game guides the avatar to move along the X and Y axis following specific paths. The avatar mimics the movement of the user's hand that performs these movements for rehabilitation. In order to use this method for the training and assessment patient’s motion, a quantified approach that uses the game-based motion for patient assessment is required. Besides simple game scores that are often used, the proposed data analysis aims to elaborate on the discrimination between pathological and healthy movement with a machine learning approach, as well as the quantification of the patient’s progress over time. For this purpose, movement and performance-related features were extracted from the leap sensor recordings and their value was explored towards characterizing the patient state and progress in detail. A dataset with multiple recordings from patients and healthy individuals was used for this purpose. All patients suffered from neuromuscular disorders. The features with the highest discriminatory value between the two groups were subsequently used to develop a set of classifiers for different sets of movements (e.g., horizontal, diagonal, vertical). A patient was left out of the classifier creation procedure and used for external validation. The models achieved high accuracy (92.13%). These results are deemed promising for the quantification of a patient’s progress.

Paper Nr: 17
Title:

Estimation of Movement Speed in Monitoring Systems based on Sensors of Multiple Types

Authors:

Jakub Wagner and Paweł Mazurek

Abstract: The research reported in this paper is related to the differentiation and fusion of measurement data in systems for healthcare-oriented unobtrusive monitoring of elderly persons. Two methods for regularised numerical differentiation – suitable for different shapes of trajectories of the monitored person’s movement – are considered. A technique for the fusion of data from sensors of different types – which involves weighting those data according to the available a priori information about the variances of errors corrupting those data – is presented. Guidelines on the usage and optimisation of that technique are provided according to the results of numerical experimentation based on synthetic data.

Paper Nr: 20
Title:

Self-Similarity Matrix of Morphological Features for Motion Data Analysis in Manufacturing Scenarios

Authors:

António Santos, João Rodrigues, Duarte Folgado, Sara Santos, Carlos Fujão and Hugo Gamboa

Abstract: There is a significant interest to evaluate the exposure that operators are subjected throughout the working day. The objective evaluation of occupational exposure with direct measurements and the need for automatic annotation of relevant events arose. Using time series retrieved from inertial sensors, this work proposes a method that is able to automatically: (1) detect anomalies, (2) segment the working cycles and (3) by means of query-by-example, identify sub segments along the working cycle. In a short summary, this technique firstly organizes the dataset provided by all inertial measurement units (IMUs) sensors placed over the dominant upper limb. After this, it retrieves a wide variety of features to an organized matrix and then calculates the respective self-similarity matrix (SSM). This method provides information by comparing each subsequence of the time series with the remaining subsequences. As the identified structures will provide information about how repetitive or anomalous is the behaviour of the data in function of time. The results show that the presented method is capable of identifying anomalies on this dataset with an accuracy of 82%, detect working cycles with a duration error of about 6% of the working cycle, and has the ability to find matches of sub-sequences of the working cycle.

Paper Nr: 31
Title:

Robust Anomaly Detection in Time Series through Variational AutoEncoders and a Local Similarity Score

Authors:

Pedro Matias, Duarte Folgado, Hugo Gamboa and André V. Carreiro

Abstract: The rise of time series data availability has demanded new techniques for its automated analysis regarding several tasks, including anomaly detection. However, even though the volume of time series data is rapidly increasing, the lack of labeled abnormal samples is still an issue, hindering the performance of most supervised anomaly detection models. In this paper, we present an unsupervised framework comprised of a Variational Autoencoder coupled with a local similarity score, which learns solely on available normal data to detect abnormalities in new data. Nonetheless, we propose two techniques to improve the results if at least some abnormal samples are available. These include a training set cleaning method for removing the influence of corrupted data on detection performance and the optimization of the detection threshold. Tests were performed in two datasets: ECG5000 and MIT-BIH Arrhythmia. Regarding the ECG5000 dataset, our framework has shown to outperform some supervised and unsupervised approaches found in the literature by achieving an AUC score of 98.79%. In the MIT-BIH dataset, the training set cleaning step removed 60% of the original training samples and improved the anomaly detection AUC score from 91.70% to 93.30%.

Paper Nr: 41
Title:

Machine Learning based Voice Analysis in Spasmodic Dysphonia: An Investigation of Most Relevant Features from Specific Vocal Tasks

Authors:

Giovanni Costantini, Pietro Di Leo, Francesco Asci, Zakarya Zarezadeh, Luca Marsili, Vito Errico, Antonio Suppa and Giovanni Saggio

Abstract: Adductor-type spasmodic dysphonia (ASD) is a task-specific speech disorder characterized by a strangled and strained voice. We have previously demonstrated that advanced voice analysis, performed with support vector machine, can objectively quantify voice impairment in dysphonic patients, also evidencing results of voice improvements due to symptomatic treatment with botulinum neurotoxin type-A injections into the vocal cords. Here, we expanded the analysis by means of three different machine learning algorithms (Support Vector Machine, Naïve Bayes and Multilayer Percept), on a cohort of 60 ASD patients, some of them also treated with botulinum neurotoxin type A therapy, and 60 age and gender-matched healthy subjects. Our analysis was based on sounds produced by speakers during the emission of /a/ and /e/ sustained vowels and a standardized sentence. As a conclusion, we report the main features with discriminatory capabilities to distinguish untreated vs. treated ASD patients vs. healthy subjects, and a comparison of the three classifiers with respect to their discriminating accuracy.

Paper Nr: 48
Title:

Performance of Monosyllabic vs Multisyllabic Diadochokinetic Exercises in Evaluating Parkinson’s Disease Hypokinetic Dysarthria from Fluency Distributions

Authors:

Pedro Gómez-Vilda, Andrés Gómez-Rodellar, Daniel Palacios-Alonso and Athanasios Tsanas

Abstract: Hypokinetic Dysarthria (HD) is a major debilitating symptom in the vast majority of people diagnosed with Parkinson's Disease (PD). It has been traditionally evaluated using diadochokinetic exercises to estimate its degree of severity, among them, the fast repetition of monosyllables as [pa], [ta], and [ka] and multisyllable sequences as [pataka], [pakata], [badaga] and others alike. However, the real efficiency of these exercises in differentiating the participant behaviour as pathological or normative has not been investigated in depth. The aim of the present work is to explore the timely responsive performance of two of these exercises (a monosyllabic [ta] vs a multisyllabic [pataka]). A method to characterize statistically syllabic and inter-syllabic interval durations in the execution of these diadochokinetic exercises, based on Kolmogorov-Smirnov approximations and Jensen-Shannon Divergence has been used to assess the efficiency of both types of exercises. The results from the evaluation of 24 gender-balanced participants (12 PD and 12 controls) show that the monosyllabic exercise does not seem to differentiate well, whereas the multisyllabic exercise has a better differentiation performance. These findings, although relatively preliminary due to the limited sample size, underline the need to carefully consider the battery of tests towards assessing HD.

Paper Nr: 49
Title:

Assessing Parkinson’s Disease Speech Signal Generalization of Clustering Results across Three Countries: Findings in the Parkinson’s Voice Initiative Study

Authors:

Athanasios Tsanas and Siddharth Arora

Abstract: Progress in exploring speech and Parkinson’s Disease (PD) has been hindered due to the use of different protocols across research labs/countries, single-site studies with relatively small numbers, and no external validation. We had recently reported on the Parkinson’s Voice Initiative (PVI), a large study where we collected 19,000+ sustained vowel phonations (control and PD groups) across seven countries, under acoustically non-controlled conditions. In this study, we explored how well findings generalize in the three English-speaking PVI cohorts (data collected in Boston, Oxford, and Toronto). We acoustically characterized each sustained vowel /a/ phonation using 307 dysphonia measures which had previously been successfully employed in speech-PD applications. We used the previously identified feature subset from the Boston cohort and explored hierarchical clustering with Ward’s linkage combined with 2D-data projections using t-distributed stochastic neighbor embedding to facilitate visual exploration of PD subgroups. Furthermore, we computed feature weights using LOGO to assess feature selection consistency towards differentiating PD from controls. Overall, findings are very consistent across the three cohorts, strongly suggesting the presence of four main PD clusters, and consistent identification of key contributing features. Collectively, these findings support the generalization of sustained vowels and robustness of the presented methodology across the English-speaking PVI cohorts.

Short Papers
Paper Nr: 4
Title:

Wavelet Correlation of Non-stationary Bursts of EEG

Authors:

S. V. Bozhokin and I. B. Suslova

Abstract: The problem of non-stationary correlation of signals recorded in various EEG channels of the human brain is considered. Each signal is represented as a sequence of bursts occurring at different times in different spectral ranges. To solve the problem of detecting the relationship of these signals, the authors introduced a new wavelet correlation function WCF. The WCF function allows you to detect the correlation of individual bursts of different EEG channels that have the same frequency, but different times of occurrence. The WCF function is built on the basis of continuous wavelet transforms of two signals taken at different points in time. For the considered recording data of two EEG channels, the burst correlations were classified. The proposed solution to the problem of correlation of EEG signals makes it possible to trace the propagation of disturbances in the cerebral cortex (traveling waves) and to reveal the synchronization of movement of evoked potentials.

Paper Nr: 5
Title:

Cross-phase Emotion Recognition using Multiple Source Domain Adaptation

Authors:

Ke-Ming Ding, Tsukasa Kimura, Ken-ichi Fukui and Masayuki Numao

Abstract: EEG signal, the brain wave, has been widely applied in detecting human emotion. Due to the human brain’s complexity, the EEG pattern varies from different individuals, leading to low cross-subject classification performance. What is more, even within the same subject, EEG data also shows diversity for the same reason. Many researchers have conducted experiments to deal with the variance between subjects by transfer learning or domain adaptation. However, most of them are still low-performance, especially when the new subject does not share generality with training samples. In this study, we examined using cross-phase data instead of cross-subject data because the discrepancy of different phase data should be smaller than that of different subjects. Different phases represent data recorded multiple times from the same subject with the same stimuli. Two neural networks are adopted to verify the effectiveness of the cross-phase domain adaptation. As a result, experiments on the public EEG dataset showed approximation level accuracy compared to the state-of-the-art method but much lower standard derivation. Moreover, multiple source domains promote accuracy in contrast to one single domain. This study helps develop a more robust and high-performance real-time EEG system by transferring knowledge from previous data phases.

Paper Nr: 8
Title:

Tragus based Vagus Nerve Stimulation for Stress Reduction

Authors:

Surej Mouli, Ramaswamy Palaniappan, Jane Ollis, Ian McLoughlin, Rahul Kanegaonkar and Sunil Arora

Abstract: Non-invasive vagus nerve stimulation is fast becoming a popular alternative treatment method for various health disorders. The authors investigated the effects of auricular vagus nerve stimulation at tragus for activating the parasympathetic nervous system to reduce stress, in light of mixed results from other studies. Stimulation frequency of 25 Hz with a pulse-width of 200 µs was administered at tragus with ECG data recorded during pre- and post-stimulation trials to investigate changes in the low-frequency (LF) and high-frequency (HF) parameters of heart rate variability (HRV). The results from five subjects demonstrate an increase in the HF component and a decrease in LF when comparing pre- and post-stimulation values, denoting that VNS stimulated more of the parasympathetic activity. The LF/HF ratio was reduced for all participants after stimulation, with an average reduction of 64.5% observed. Overall, this study has indicated the feasibility of using tragus as a stimulation site to stimulate the vagus nerve; tragus being easier to administrate than many alternative sites while still being effective for stress reduction.

Paper Nr: 13
Title:

Convolutional Neural Networks for Quantitative Prediction of Different Organic Materials using Near-Infrared Spectrum

Authors:

Tegegn D. Delelegn, Italo F. Zoppis, Sara Manzoni, Cezar Sas and Edoardo Lotti

Abstract: Advances in Near-infrared (NIR) spectroscopy technology led to an increase of interest in its applications in various industries due to its powerful non-destructive quantization tool. In this work, we used a one-dimensional CNN to determine simultaneously quantities of organic materials in a mixture using their NIR infrared spectra. The coefficient of determination (R2) and the root mean square error (RMSE) is used to test the performance of the model. We used six materials to make pairwise combinations with distinct quantities of each pair. We obtained 13 different pairwise mixtures, afterward, their near-infrared spectrum profiles is extracted. The model predicted for each mixture their percentage of composition with a result of 0.9955 R2 and RMSE 0.0199. Furthermore, we examined the performance of our model when predicting unseen composition percentages with unseen mixtures. To do so, two scenarios are carried out by filtering the training and testing set: the first one where we test on unseen composition percentage (UP) of mixtures, and the second one where we test on unseen composition percentage of unseen mixtures (UPM). The model achieved an R2 of 0.947 and 0.627 scores respectively for UP and UPM.

Paper Nr: 15
Title:

Feature Extraction for Stress Detection in Electrodermal Activity

Authors:

Erika Lutin, Ryuga Hashimoto, Walter De Raedt and Chris Van Hoof

Abstract: Electrodermal activity (EDA) is a sensitive measure for changes in the sympathetic system, reflecting emotional and cognitive states such as stress. There is, however, inconsistency in the recommendations on which features to extract. In this study, we brought together different feature extraction methods: trough-to-peak features, decomposition-based features, frequency features and time-frequency features. Regarding the decomposition analysis, three different applications were used: Ledalab, cvxEDA and sparsEDA. A total of forty-seven features was extracted from a previously collected dataset. This dataset included twenty participants performing three different stress tasks. A Support Vector Machine (SVM) classifier was built in a Leave-One-Subject-Out Cross Validation (LOOCV) set-up with feature selection within the LOOCV loop. Three features were consistently selected over all participants: 1) the number of responses in the driver function generated by Ledalab and 2) by sparsEDA and 3) a time-frequency feature, previously described as TVSymp. The classifier obtained an accuracy of 88.52%, a sensitivity of 72.50% and a specificity of 93.65%. This research shows that EDA can be successfully used in stress detection, without the addition of any other physiological signals. The classifier, built with the most recent feature extraction methods in literature, was found to outperform previous classification attempts.

Paper Nr: 16
Title:

Comparison of Convolutional and Recurrent Neural Networks for the P300 Detection

Authors:

Lukáš Vařeka

Abstract: Single-trial classification of the P300 component is a difficult task because of the low signal to noise ratio. However, its application to brain-computer interface development can significantly improve the usability of these systems. This paper presents a comparison of baseline linear discriminant analysis (LDA) with convolutional (CNN) and recurrent neural networks (RNN) for the P300 classification. The experiments were based on a large multi-subject publicly available dataset of school-age children. Several hyperparameter choices were experimentally investigated and discussed. The presented CNN slightly outperformed both RNN and baseline LDA classifier (the accuracy of 63.2 % vs. 61.3 % and 62.8 %). The differences were most pronounced in precision and recall. Implications of the results and proposals for future work, e.g., stacked CNN–LSTM, are discussed.

Paper Nr: 18
Title:

Surface EMG-based Profiling and Fatigue Analysis of the Biceps Brachii Muscle of Cricket Bowlers

Authors:

Muhammad U. Rizwan, Nadeem A. Khan, Rushda B. Ahmad and Muneeb Ijaz

Abstract: Cricket bowling action is a complex repetitive task involving multiple muscles. In this paper we present a protocol to analyse accumulated localized fatigue in muscles during cricket bowling action. Biceps Brachii (BB) muscle in case of fast delivery for a novice player is analysed to illustrate the methodology. Synchronized video recording with the surface EMG signal was captured from the medial position of the BB muscle to enable segmentation of the EMG signal in six intervals corresponding to the six phases of the bowling action. This enables study of the activation pattern of the muscle along with the fatigue trend during bowling. Both integrated EMG and Mean Power frequency (MPF) are used as measures to analyse fatigue. Though we have plotted the trends for a single muscle, a similar exercise should be repeated for all important muscles involved. Analysing localized fatigue in individual muscles is important for injury prevention as well as player performance development. It can help to see how individual muscle fatigue contributes in declining performance during cricket bowling. Such an analysis can also be used to support minimum bowling overs and suitable inter-over breaks for a specific bowler with regard to injury prevention and optimal performance.

Paper Nr: 19
Title:

Mental Workload Estimation using Wireless EEG Signals

Authors:

Quadri Adewale and George Panoutsos

Abstract: Previous studies have demonstrated the applicability of electroencephalogram (EEG) in estimating mental workload. However, developing reliable models for cross-task, cross-subject and cross-session classifications of workload remains a challenge. In this study, we used a wireless Emotiv EPOC headset to evaluate workload in eight subjects and two mental tasks, namely n-back, and arithmetic tasks. 0-back and 2-back tasks, and 1-digit and 3-digit additions were employed as low and high workloads in the n-back and arithmetic tasks, respectively. Using power spectral density as features, a signal processing and feature extraction framework was developed to classify workload levels. Within-session accuracies of 98.5% and 95.5% were achieved in the n-back and arithmetic tasks, respectively. To facilitate real-time estimation of workload, a fast domain adaptation technique was applied to achieve a cross-task accuracy of 68.6%. Similarly, we obtained accuracies of 80.5% and 76.6% across sessions, and 74.4% and 64.1% across subjects, in n-back and arithmetic tasks, respectively. Although the number of participants is limited, this framework generalised well across subjects and tasks, and provides a promising approach towards developing subject and task-independent models. It also shows the feasibility of using a consumer-level wireless EEG headset in cognitive monitoring for real-time estimation of workload in practice.

Paper Nr: 21
Title:

Noise-resilient Automatic Interpretation of Holter ECG Recordings

Authors:

Egorov Konstantin, Sokolova Elena, Avetisian Manvel and Tuzhilin Alexander

Abstract: Holter monitoring, a long-term ECG recording (24-hours and more), contains a large amount of valuable diagnostic information about the patient. Its interpretation becomes a difficult and time-consuming task for the doctor who analyzes them because every heartbeat needs to be classified, thus requiring highly accurate methods for automatic interpretation. In this paper, we present a three-stage process for analysing Holter recordings with robustness to noisy signal. First stage is a segmentation neural network (NN) with encoder-decoder architecture which detects positions of heartbeats. Second stage is a classification NN which will classify heartbeats as wide or narrow. Third stage in gradient boosting decision trees (GBDT) on top of NN features that incorporates patient-wise features and further increases performance of our approach. As a part of this work we acquired 5095 Holter recordings of patients annotated by an experienced cardiologist. A committee of three cardiologists served as a ground truth annotators for the 291 examples in the test set. We show that the proposed method outperforms the selected baselines, including two commercial-grade software packages and some methods previously published in the literature.

Paper Nr: 22
Title:

Feature Space Reduction for Human Activity Recognition based on Multi-channel Biosignals

Authors:

Yale Hartmann, Hui Liu and Tanja Schultz

Abstract: In this paper, we study the effect of Feature Space Reduction for the task of Human Activity Recognition (HAR). For this purpose, we investigate a Linear Discriminant Analysis (LDA) trained with Hidden Markov Models (HMMs) force-aligned targets. HAR is a typical application of machine learning, which includes finding a lower-dimensional representation of sequential data to address the curse of dimensionality. This paper uses three datasets (CSL19, UniMiB, and CSL18), which contain data recordings from humans performing more than 16 everyday activities. Data were recorded with wearable sensors integrated into two devices, a knee bandage and a smartphone. First, early-fusion baselines are trained, utilizing an HMM-based approach with Gaussian Mixture Models to model the emission probabilities. Then, recognizers with feature space reduction based on stacking combined with an LDA are evaluated and compared against the baseline. Experimental results show that feature space reduction improves balanced accuracy by ten percentage points on the UniMiB and seven points on the CSL18 datasets while remaining the same on the CSL19 dataset. The best recognizers achieve 93.7 ± 1.4% (CSL19), 69.5 ± 8.1% (UniMiB), and 70.6 ± 6.0% (CSL18) balanced accuracy in a leave-one-person-out cross-validation.

Paper Nr: 26
Title:

Estimation of Affective State based on Keystroke and Typing Vibration during Computer-Mediated Communication

Authors:

Kei Hasegawa, Hikaru Miyamoto, Yuki Ashida, Yuki Ban, Rui Fukui, Masahiro Inazawa and Shin’ichi Warisawa

Abstract: In recent years, the use of computer-mediated communication (CMC), that is, communication among people through computers, has increased. Knowing the message sender’s affective state is essential for understanding the contents of the message correctly. However, it is difficult to interpret this state because of the nonavailability of nonverbal information from the sender during CMC. Although attempts have been performed to estimate affective state, there is a challenge of high measurement load. In this paper, we propose an estimation of valence and arousal using keyboard input and typing vibration information as a method to estimate the sender’s affective state with a low measurement load during CMC. We conducted experiments to obtain keyboard input and typing vibration information for estimating valence and arousal. This estimation was performed by extracting features from the information using a support vector machine, and cross-validation was conducted to verify our method. Therefore, the valence and arousal were estimated at accuracies of 69.8% and 71.1%, respectively, for unlearned participants’ data.

Paper Nr: 28
Title:

Automatic Segmentation of Mammary Tissue using Computer Simulations of Breast Phantoms and Deep-learning Techniques

Authors:

Lucca R. Peregrino, Jordy V. Gomes, Thaís G. do Rêgo, Yuri M. Barbosa, Telmo S. Filho, Andrew A. Maidment and Bruno Barufaldi

Abstract: Digital breast tomosynthesis (DBT) has rapidly emerged for screening mammography to improve cancer detection. Segmentation of dense tissue plays an important role in breast imaging applications to estimate cancer risk. However, the current segmentation methods do not guarantee an ideal ground-truth in clinical practice. Computer simulations provide ground-truth that enables the development of convolutional neural network (CNN) applications designed for image segmentation. This study aims to train a CNN model to segment dense tissue in DBT images simulated using anthropomorphic phantoms. The phantom images were simulated based on clinical settings of a DBT system. A U-Net, a CNN model, was trained with 2,880 images using a slice-wise approach. The U-Net performance was evaluated in terms of percent of density in the central slice and volumetric breast density in the medio-lateral slices. Our results show that the U-Net can segment dense tissue from DBT images with overall loss, accuracy, and intersection over union of 0.27, 0.93, and 0.62 in the central slices, and 0.32, 0.92, and 0.54 in the medio-lateral slices, respectively. These preliminary results allow us to explore the use of CNN architectures to segment dense tissue in clinical images, which is a highly complex task in screening with DBT.

Paper Nr: 30
Title:

A Wavelet Scattering Convolutional Network for Magnetic Resonance Spectroscopy Signal Quantitation

Authors:

Amirmohammad Shamaei, Jana Starčuková and Zenon Starčuk Jr.

Abstract: Magnetic resonance spectroscopy (MRS) can provide quantitative information about local metabolite concentrations in living tissues, but in practice the quantification can be difficult. Recently deep learning (DL) has been used for quantification of MRS signals in the frequency domain, and DL combined with time-frequency analysis for artefact detection in MRS. The networks most widely used in previous studies were Convolutional Neural Networks (CNN). Nonetheless, the optimal architecture and hyper-parameters of the CNN for MRS are not well understood; CNN has no knowledge about the nature of the MRS signal and its training is computationally expensive. On the other hand, Wavelet Scattering Convolutional Network (WSCN) is well-understood and computationally cheap. In this study, we found that a wavelet scattering network could hopefully be also used for metabolite quantification. We showed that a WSCN could yield results more robust than QUEST (one of quantitation methods based on model fitting) and the same as a CNN while being faster. We used wavelet scattering transform to extract features from the MRS signal, and a superficial neural network implementation to predict metabolite concentrations. Effects of phase, noise, and macromolecules variation on the WSCN estimation accuracy were also investigated.

Paper Nr: 32
Title:

Eppur si muove: Formant Dynamics is Relevant for The Study of Speech Aging Effects

Authors:

Luciana Albuquerque, Catarina Oliveira, António Teixeira and Daniela Figueiredo

Abstract: The evidence have shown that speech change with age and the automatic speech recognition systems needs adaptation to older voices. Most of the acoustic studies about the age effects on speech production have focused on static approaches to obtain the vowel formants. However, vowel formant dynamics may also be important to characterize vowel quality and the age related changes. In this position paper the authors argue for the need to increase the use of dynamic information in acoustic studies. Among the main arguments, we can state that: speech is inherently dynamic; dynamic vowel formants improve the classification of vowels and dialects and play an important role in vowel perception; nowadays better tools allow to go beyond analysis of snapshots.

Paper Nr: 34
Title:

Estimation of Chronic Stress by Measuring Sympathetic Sedation Time

Authors:

Masahiro Inazawa, Yuki Ban, Naoki Tateyama and Shin’ichi Warisawa

Abstract: Intermittent exposure to stressors disrupts the negative feedback mechanism of cortisol toward corticotropin-releasing hormones. In this study, this condition is referred to as chronic stress. Chronic stress causes a variety of recurring, long-term, incurable illnesses, such as major depression. Therefore, it is important to understand chronic stress on a daily basis. We propose a chronic stress estimation method using sympathetic sedation time measurements as a non-invasive, short-time, and highly accurate method. This method determines the degree of chronic stress according to the length of time until the sympathetic activity subsides after stressor loading. To verify the feasibility of the proposed method, we conducted an experiment comparing the sympathetic sedation times among a healthy group, middle group, and chronic stress group classified by the Quick Inventory of Depressive Symptomatology. We calculated sympathetic sedation time from the trend of change in RRV at calm after stressor loading due to a two-back task. As a result of the experiment, which consisted of nine participants, the sympathetic sedation time in the chronic stress group was longer than in the healthy and middle groups, supporting the feasibility of this method.

Paper Nr: 35
Title:

Prototype Reduction on sEMG Data for Instance-based Gesture Learning towards Real-time Prosthetic Control

Authors:

Tim Sziburis, Markus Nowak and Davide Brunelli

Abstract: Current systems of electromyographic prostheses are controlled by machine learning techniques for gesture detection. Instance-based learning showed promising results concerning classification accuracy and robustness without explicit model training. However, it suffers from high computational demands in the prediction phase, which can be problematic in real-time scenarios. This paper aims at combining such learning schemes with the concept of prototype reduction to decrease the amount of data processed in each prediction step. First, a suitability assessment of state-of-research reduction algorithms is conducted. This is followed by a practical feasibility analysis of the approach. For this purpose, several datasets of signal classes from exerting specific gestures are captured with an eight-channel EMG armband. Based on the recorded data, prototype reduction algorithms are comparatively applied. The dataset reduction is characterized by the time needed for reduction as well as the possible data reduction rate. The classification accuracy when using the reduced set in cross-validation is analyzed with an exemplary kNN classifier. While showing promising values in reduction time as well as excellent classification accuracy, a reduction rate of over 99% can be achieved in all tested gesture configurations. The reduction algorithms LVQ3 and DSM turn out to be particularly convenient.

Paper Nr: 36
Title:

On the Optimal Strategy for Tackling Head Motion in fMRI Data

Authors:

Júlia F. Soares, Rodolfo Abreu, Ana C. Lima, Sónia Batista, Lívia Sousa, Miguel Castelo-Branco and João V. Duarte

Abstract: Head motion critically hampers the quality of functional magnetic resonance imaging (fMRI) data, with several methods for its correction being already available in the literature. Head shifts are usually corrected by realigning all functional volumes with relation to a reference volume using affine transformations, from which the estimated motion parameters (MPs) can be additionally regressed out from fMRI data. However, a consensus regarding the number of MPs to regress has not been achieved yet. More critically, abrupt head motion induces the so-called motion outliers in the data, which cannot be accounted for by affine transformations. Two common approaches are widely used to tackle this type of motion, namely modelling strategies such as censoring, and volume interpolation. However, a direct comparison between strategies to tackle motion outliers has not been performed so far. Importantly, to our knowledge no study has focused on determining the extent at which the effects of different head motion correction methods differ between groups in clinical studies. This is particularly relevant in task-related functional connectivity fMRI studies, which are rapidly increasing in clinical research. In this study, we started by determining the optimal number of MPs (between 6 and 24) to be regressed out from fMRI data collected from 8 participants (4 patients with Multiple Sclerosis and 4 healthy controls) performing a perceptual decision-making task. Then we tested motion censoring and volume interpolation for correcting motion outliers, using FD and DVARS metrics to detect the outlier volumes. We found that task-specific activated brain regions were detected with higher sensitivity when using 6 MPs relatively to using 24 MPs. As for the correction of motion outliers, our results suggest that volume interpolation is the best method to use, however more data and external validation is needed to achieve a definite conclusion. Importantly, the performance of motion correction algorithms was irrespective of the subject group (patients and healthy participants). Our results pave the way towards finding an optimal motion correction strategy, which is required to improve the accuracy of fMRI analyses in healthy and patient populations and are an encouragement to test comprehensively different approaches.

Paper Nr: 38
Title:

FRADE: Pervasive Platform for Fall Risk Assessment, Prevention and Fall Detection

Authors:

Joana Silva, Nuno Cardoso, Jorge Ribeiro, Alberto Carvalho, Mariana Pereira, Fernando Ricaldoni, Carlos Resende and João Oliveira

Abstract: The ageing of the global population has an impact on the elderly quality of life, as the reduced mobility and balance contribute to the increasing of falls. Fall detection solutions can trigger emergency alerts and reduce the negative effect of falls. Fall risk assessment strategies can help to early identify fall risk factors and tailor strategies to revert those risk factors, by means of fall prevention exercises. However, most technological solutions do not simultaneously address these three aspects of the falls management cycle. FRADE platform will allow to pervasively detect falls using a wearable device, that can also be used to monitor fall risk assessment tests and recommended individual exercises, that can be performed at home with a tablet and two wearables.

Paper Nr: 39
Title:

Inertial-based Gait Analysis Applied to Patients with Parkinson Disease

Authors:

Joana Sousa, Joana Silva, Ricardo Leonardo, Hugo Gamboa and Josefa Domingos

Abstract: People with Parkinson’s disease have a high incidence of falls due to motor difficulties. Recent studies have shown that PD patients can receive benefit from motor therapy based on cueing and feedback. This study describes a system based on a foot-mounted IMU for the calculation of gait parameters applied to different datasets of healthy elderly people, geriatric patients and patients with PD, in order to integrate it into a real-time acquisition system with application for tactile cueing. This system is divided into different steps: the identification of gait cycles and their events, the estimation of the path of the foot, which includes the estimation of the orientation of the foot, the application of methods to correct the error derived from the double integration of acceleration such as ZUPT, and finally the estimation of the different gait metrics. The results show that the algorithm developed is an accurate method for stride segmentation and is considered adequate to assess the gait metrics for gait evaluation of patients with motor difficulties.

Paper Nr: 40
Title:

Evaluating Synthetic Speech Workload with Oculo-motor Indices: Preliminary Observations for Japanese Speech

Authors:

Mateusz Dubiel, Minoru Nakayama and Xin Wang

Abstract: Pupillometry has recently been introduced as a method to evaluate cognitive workload of synthetic speech. Prior research conducted on English speech indicates that in noisy listening conditions, pupil dilation is significantly higher for synthetic speech as compared to natural speech. In a lab-based listening experiment, we evaluated participants’ (n=16) pupil responses to Japanese speech (natural vs. synthetic) at three different signal-to-noise levels (-1dB, -3dB and -5dB). Our research expands on previous work by evaluating pupillary responses both in terms of temporal changes in pupil size and degree of pupil oscillations. We observe statistically significant differences in pupil sizes at the recall stage between each type of speech. For pupil oscillations, we register statistically significant differences in frequency power spectrum densities (PSDs). Our investigation proposes an expansion of the current synthetic speech evaluation methods that are based on pupillary responses and outlines possible avenues for future research that arise from the findings of this work.

Paper Nr: 45
Title:

Convolution-based Soma Counting Algorithm for Confocal Microscopy Image Stacks

Authors:

Shih-Ting Huang, Yue Jiang and Hao-Chiang Shao

Abstract: To facilitate brain research, scientists need to identify factors that can promote or suppress neural cell differentiation mechanisms. Accordingly, the way to recognize, segment, and count developing neural cells within a microscope image stack becomes a fundamental yet considerable issue. However, it is currently not feasible to develop a DCNN (deep convolutional neural network) based segmentation algorithm for confocal fluorescence image stacks because of the lack of manual-annotated segmentation ground truth. Also, such tasks traditionally require meticulous manual preprocessing steps, and such manual steps make the results unstable even with software support like ImageJ. To solve this problem, we propose in this paper a convolution-based algorithm for cell recognizing and counting. The proposed method is computationally efficient and nearly parameter-free. For a 1024×1024×70 two-channel image volume containing about 100 developing neuron cells, our method can finish the recognition and counting tasks within 250 seconds with a standard deviation smaller than 4 comparing with manual cell-counting results

Paper Nr: 46
Title:

Automatic Emotion Recognition from DEMoS Corpus by Machine Learning Analysis of Selected Vocal Features

Authors:

Giovanni Costantini, E. Parada-Cabaleiro and Daniele Casali

Abstract: Although Speech Emotion Recognition (SER) has become a major area of research in affective computing, the automatic identification of emotions in some specific languages, such as Italian, is still under-investigated. In this regard, we assess how different machine learning methods for SER can be applied in the identification of emotions in Italian language. In agreement with studies that criticize the use of acted emotions in SER, we considered DEMoS, a new database in Italian built through mood induction procedures. The corpus consists of 9365 spoken utterances produced by 68 Italian native speakers (23 females, 45 males) in a variety of emotional states. Experiments were carried out for female and male separately, considering for each a specific feature set. The two feature sets were selected by applying Correlation-based Feature Selection from the INTERSPEECH 2013 ComParE Challenge feature set. For the classification process, we used Support Vector Machine. Confirming previous work, our research outcomes show that the basic emotions anger and sadness are the best identified, while others more ambiguous, such as surprise, are worse. Our work shows that traditional machine learning methods for SER can be also applied in the recognition of an under-investigating language, such as Italian, obtaining competitive results.

Paper Nr: 47
Title:

The Relationship between Psychological Workload and Oculomotor Indices under Visual Search Task Execution

Authors:

Tomomi Okano and Minoru Nakayama

Abstract: In this paper, we have focused especially on microsaccade and pupil diameter to extract relationships with psychological workload. We measured how these oculomotor feature values changes to 10 subjects when executing visual search tasks containing psychological workload. To evaluate the amount of psychological workload, we used a systematic evaluation index, NASA-TLX and analyzed by combining pupil movements with answer rate and difficulty of both tasks. As a result, we have discovered that by the difference of psychological workload and 2 experimental conditions, microsaccade frequency and task performance changes.

Paper Nr: 3
Title:

Deep Learning Solution for Pathological Voice Detection using LSTM-based Autoencoder Hybrid with Multi-Task Learning

Authors:

Dávid Sztahó, Kiss Gábor and Tulics M. Gábriel

Abstract: In this paper, a deep learning approach is introduced to detect pathological voice disorders from continuous speech. Speech as bio-signal is getting more and more attention as a discriminant for different diseases. To exploit information in speech, a long-short term memory (LSTM) autoencoder hybrid with multi-task learning solution is proposed with spectrogram as input feature. Different speech databases (voice disorders, depression, Parkinson’s disease) are applied as evaluation datasets. Applicability of the method is demonstrated by obtaining accuracies 85% for Parkinson’s disease, 86% for dysphonia, and 90% for depression on test datasets. The advantage of this method is that it is fully data-driven, in the sense that it does not require special acoustic-phonetic preprocessing separately for the types of disease to be recognized. We believe that the applied method in this article can be used to other diseases as well and can be used for other languages also.

Paper Nr: 6
Title:

Clustering Pathologic Voice with Kohonen SOM and Hierarchical Clustering

Authors:

Alessa A. de Oliveira, Maria E. Dajer and João P. Teixeira

Abstract: The main purpose of clustering voice pathologies is the attempt to form large groups of subjects with similar pathologies to be used with Deep-Learning. This paper focuses on applying Kohonen's Self-Organizing Maps and Hierarchical Clustering to investigate how these methods behave in the clustering procedure of voice samples by means of the parameters absolute jitter, relative jitter, absolute shimmer, relative shimmer, HNR, NHR and Autocorrelation. For this, a comparison is made between the speech samples of the Control group of subjects, the Hyper-functional Dysphonia and Vocal Folds Paralysis pathologies groups of subjects. As a result, the dataset was divided in two clusters, with no distinction between the pre-defined groups of pathologies. The result is aligned with previous result using statistical analysis.

Paper Nr: 23
Title:

Segmented ECG Bio Identification using Fréchet Mean Distance and Feature Matrices of Fiducial QRS Features

Authors:

Abdullah Biran and Aleksandar Jeremic

Abstract: In this paper, we present a new segmented based method for human identification using Fréchet distances and the characteristics of the lag-feature matrices of six fiducial based QRS features. We examined the applicability of our methodology on 124 ECG records of 62 subjects from the publicly available ECG ID data base. Our experiments show that the Fréchet distance can identify majority of the subjects (44 individuals) using the feature matrix of QRS segment lagged by one beat with an identification accuracy ranging from 80% to 100%. Our preliminary results indicate that identifying humans using segmented approaches can be potentially useful.

Paper Nr: 24
Title:

A Multi-spot Murmur Sound Detection Algorithm and Its Application to a Pediatric and Neonate Population

Authors:

Marisa Oliveira, Jorge Oliveira, Rui Camacho and Carlos Ferreira

Abstract: Cardiovascular diseases are one of the leading causes of death in the world. In low income countries, heart auscultation is of capital importance since it is an efficient and low cost method to monitor the heart. In this paper, we propose a multi-spot system that aims to detect cardiac anomalies and to support a diagnosis in remote areas with limited heath care response. Our proposed solutions exploits data collected from the four main auscultation spots: Mitral, Pulmonary, Tricuspid and Aorta in a asynchronous way. From the several multi-spot systems implemented, the best results were obtained using a bi-modal system that only processes the Mitral and the Pulmonary spot simultaneously. Using these two spots we have achieved an accuracy between 85.7% (smallest value, using ANN) and the best value of 91.4% (obtained with a logistic regression algorithm). Taking into a account the pediatric population and the incident cardiac pathologies, it happens to be the spots where the observed murmurs were most audible. We have also find out that when using four auscultation spots, the choice of the algorithm is of secondary priority, which does not seem to be the case for a single auscultation spot system. With one single auscultation we have an average of 4% of difference between the results obtained with the algorithms and with four auscultation spots we have a smaller average of 2.1%.

Paper Nr: 27
Title:

Classification of Myoelectric Surface Signals of Hand Movements using Supervised Learning Techniques

Authors:

Marisol G. Flores, Cristian D. Álamo and Juan M. Medina

Abstract: This work presents a comparative study of techniques to classify four hand movements (flexion, extension, opening and closure) using myoelectric signals measured at the forearm in two separate channels: the brachioradialis and the flexor carpi ulnaris (FCU) muscle. The process of signal acquisition is described, as well as signal normalization, hybrid feature extraction and classification using two supervised learning techniques; i.e., backpropagation and support vector machines. The classifiers were trained using the raw data from the input signal. It was verified that the accuracy of the classification is improved by feature extraction up to 2.25%, yielding a successful average classification rate of 91.00%.

Paper Nr: 29
Title:

Dispersion Entropy: A Measure of Electrohysterographic Complexity for Preterm Labor Discrimination

Authors:

Félix Nieto-del-Amor, Yiyao Ye-Lin, Javier Garcia-Casado, A. Diaz-Martinez, María G. Martínez, R. Monfort-Ortiz and Gema Prats-Boluda

Abstract: Although preterm labor is a major cause of neonatal death and often leaves health sequels in the survivors, there are no accurate and reliable clinical tools for preterm labor prediction. The Electrohysterogram (EHG) has arisen as a promising alternative that provides relevant information on uterine activity that could be useful in predicting preterm labor. In this work, we optimized and assessed the performance of the Dispersion Entropy (DispEn) metric and compared it to conventional Sample Entropy (SampEn) in EHG recordings to discriminate term from preterm deliveries. For this, we used the two public databases TPEHG and TPEHGT DS of EHG recordings collected from women during regular checkups. The 10th, 50th and 90th percentiles of entropy metrics were computed on whole (WBW) and fast wave high (FWH) EHG bandwidths, sweeping the DispEn and SampEn internal parameters to optimize term/preterm discrimination. The results revealed that for both the FWH and WBW bandwidths the best separability was reached when computing the 10th percentile, achieving a p-value (0.00007) for DispEn in FWH, c = 7 and m = 2, associated with lower complexity preterm deliveries, indicating that DispEn is a promising parameter for preterm labor prediction.

Paper Nr: 33
Title:

Separation Method of Atrial Fibrillation Classes with High Order Statistics and Classification using Machine Learning

Authors:

Luís Fillype da Silva, Jonathan A. Queiroz, Caroline Vanessa, Allan K. Barros, Gean C. Lopes and Letícia Cabral

Abstract: The electrocardiogram (ECG) is an exam that presents a graphical representation of the electrical activity of the heart. Through it, it is possible to observe the rhythm of heart beats, the number of beats per minute, in addition to enabling the diagnosis of various arrhythmias. This article aims to develop a classification model based on the beats of three groups of individuals: with atrial fibrillation, intra-atrial fibrillation and normal sinus rhythm. The methodology of extraction of characteristics based and adapted to classify Atrial Fibrillation and its subtype, Intracardiac Atrial Fibrillation. The classifications were carried out in three-dimensional space in two stages: with the application of Principal Component Analysis (PCA) and without application of it, through Artificial Neural Networks (ANN), Support Vector Machines (SVM) and K-nearest Neighbors (KNN), obtaining accuracy of 93% to 99%.

Paper Nr: 37
Title:

CIBA: Continuous Interruption-free Brain Authentication

Authors:

Florian Gondesen and Dieter Gollmann

Abstract: The performance of contemporary biometrics systems based on electroencephalography (EEG) suffers from a low signal to noise ratio due to the properties of the human EEG and the measurement on the scalp. There is a trade-off between accuracy and the time required for data acquisition. Additional time is needed to mount an EEG headset so that authentication requires several minutes, rendering it not very usable for most scenarios. In a scenario where an EEG headset is already worn for a different purpose, the setup time can be neglected and the time for data acquisition may be extended if it does not interfere with the subject’s actual task, allowing continuous authentication. However, most proposed EEG-based authentication systems require the user to perform a certain task during data acquisition, distracting the user from the actual task. We conceptualize an EEG-based continuous authentication scheme that does not require the user to perform a task in addition to working at a screen. We propose two approaches based on well known brain responses, SSVEP and ERP.

Paper Nr: 44
Title:

Separating Local and Propagated Contributors to the Behnke-fried Microelectrode Recordings

Authors:

P. Jurczynski, S. L. Cam, B. Rossion and R. Ranta

Abstract: In electrophysiological measurements, a recording electrode is located in an electric field and has a certain electric potential value. Each measured signal is a potential difference between two electrodes, a measuring electrode and a so-called reference electrode. This reference electrode is not located at infinity and outside of any electric field, thus its electric potential is found in the measured signals. In order to isolate the local activity of the recorded structure, it is necessary to understand the relationships between the different contributors at the electrode level and separate these different activities. In this paper, we focus on the particular setup of Behnke-Fried micro-electrodes. We propose to adapt a previous re-referencing method for separating local and distant propagated activities taking into account the non-stationarity of the signals and the particular geometry of these microelectrodes. We demonstrate, on realistically simulated signals, that the new re-referencing procedure improves the preprocessing of these signals and might help deeper interpretation.