BIOSIGNALS 2024 Abstracts

Full Papers Short Papers
Full Papers
Paper Nr: 15
Title:

Fusion of Machine Learning and Threshold-Based Approaches for Fall Detection in Healthcare Using Inertial Sensors
Authors:

Ya Wang, Peiman A. Sarvari and Djamel Khadraoui
Abstract: In the healthcare sector, specifically for elderly care, accurate and efficient fall detection is crucial. We present an advanced fall detection methodology tailored for wearable systems. Our approach blends threshold-based screening with machine learning models like Support Vector Machine, K-Nearest Neighbors, Decision Tree, Random Forest, and XGBoost. Utilizing 65 features extracted from the gyroscope and accelerometer data from Inertial Measurement Units, our method addresses the class imbalance often found between Activities of Daily Living and actual fall events. Threshold-based pre-screening serves to mitigate the class imbalance of the fall dataset, making the subsequent machine-learning classification more effective. Validation on two open-source IMU datasets, Sisfall and FallAllD, achieving high accuracy rates of 99.55%, 99.68% (wrist), 99.76% (waist), and 99.52% (neck), shows our model surpassing existing solutions in detection accuracy. Furthermore, our strategic feature extraction not only enhances the model’s performance but also allows for a fourfold reduction by using the 15 most important features in data transmission without sacrificing accuracy. These findings underscore the efficiency and potential of our methodology, indicating that wearables can indeed be powerful tools for high-precision fall detection with minimal data overhead.
Download
Paper Nr: 34
Title:

A Comparison of Recurrent and Convolutional Deep Learning Architectures for EEG Seizure Forecasting
Authors:

Sina Shafiezadeh, Marco Pozza and Alberto Testolin
Abstract: Many research efforts are being spent to discover predictive markers of seizures, which would allow to build forecasting systems that could mitigate the risk of injuries and clinical complications in epileptic patients. Although electroencephalography (EEG) is the most widely used tool to monitor abnormal brain electrical activity, no commercial devices can reliably anticipate seizures from EEG signal analysis at present. Recent advances in Artificial Intelligence, particularly deep learning algorithms, show promise in enhancing EEG classifier forecasting accuracy by automatically extracting relevant spatio-temporal features from EEG recordings. In this study, we systematically compare the predictive accuracy of two leading deep learning architectures: recurrent models based on Long Short-Term Memory networks (LSTMs) and Convolutional Neural Networks (CNNs). To this aim, we consider a data set of long-term, continuous multi-channel EEG recordings collected from 29 epileptic patients using a standard set of 20 channels. Our results demonstrate the superior performance of deep learning algorithms, which can achieve up to 99% accuracy, sensitivity, and specificity compared to more traditional machine learning approaches, which settle around 75% in all evaluation metrics. Our results also show that giving as input the recordings from all electrodes allows to exploit useful channel correlations to learn more robust predictive features, compared to convolutional models that treat each channel independently. We conclude that deep learning architectures hold promise for enhancing the diagnosis and prediction of epileptic seizures, offering potential benefits to those affected by such invalidating neurological conditions.
Download
Paper Nr: 66
Title:

Classification of Fine-ADL Using sEMG Signals Under Different Measurement Conditions
Authors:

Surya Naidu, Anish Turlapaty and Vidya Sagar
Abstract: Most studies on surface electromyography (sEMG) related to finger activities have concentrated on grips, grasps and general arm movements without any emphasis on the correlation of body postures and hand positions on the finger-centric activities. The main objective of the new dataset is to investigate activities of daily living (ADL) needing focus on finer motor control in diverse measurement conditions. In this paper, we present a novel dataset of finger-centric activities of daily living comprising 5-channel sEMG signals collected under different body postures and hand positions. The dataset encompasses sEMG signals acquired from 10 subjects, performing 10 distinct fine-ADLs in various body postures and hand positions. Further, a machine learning framework for classification of the fine-ADL is developed as follows. Each signal is segmented into 250ms windows and Time Domain (TD), Frequency Domain (FD), Wavelet Domain (WD) and Eigenvalues features are extracted. Four classifier frameworks using the features are implemented for the analyses. The results reveal that a hybrid CNN Bi-LSTM achieves an average test accuracy of 76.85%, followed by a 5-layered fully connected neural network (FCNN) with 72.42%, in aggregate scenario. An average subject-wise test accuracy of 88% is achieved by the FCNN across all body postures and hand positions combined. Most importantly, the CNN Bi-LSTM, enhances subject-wise classification by an average test accuracy of 27 .47% than the FCNN under varying body postures. Dependencies of the test accuracy on measurement conditions are analyzed. Stand body posture is found to be the easiest to classify in Aggregate scenario, whereas Folded Knees was the most difficult to classify. An increase in test accuracy with an increase in training data is observed body postures combinations analysis.
Download
Paper Nr: 71
Title:

Contactless Camera-Based Detection of Oxygen Desaturation Events and ODI Estimation During Sleep in SAS Patients
Authors:

Belmin Alić, Samuel Tauber, Reinhard Viga, Christian Wiede and Karsten Seidl
Abstract: Recurrent nocturnal breathing cessation leads to the reduction of the blood oxygen level and eventually to oxygen desaturation. Oxygen desaturation events are traditionally detected during a polysomnography in a sleep laboratory. In this work, a contactless camera-based oxygen desaturation detection and oxygen desatu-ration index (ODI) estimation method based on the analysis of multispectral videos is proposed. The method is based on the extraction and analysis of remote photoplethysmography (rPPG) signals at wavelengths of 780 nm and 940 nm from the forehead and a breath temperature signal via far-infrared (FIR) thermography from the subnasal region. A manual feature extraction is designed to extract relevant medical and physiological parameters out of the aforementioned signals in order to design a Feed-Forward Neural Network (FFNN)-based classifier, which classifies between periods with and without desaturation events. For the evaluation of the proposed method, a patient dataset involving 23 symptomatic sleep apnea patients is collected. The classification accuracy between desaturation events and periods without a desaturation based on the leave-one-patient-out cross-validation (LOPOCV) metric is 95.4 %. The ODI stage estimation resulted in a correct estimation in 22 out of 23 patients for a two-stage ODI classification and in a correct estimation in 21 out of 23 patients for a four-stage ODI classification.
Download
Paper Nr: 98
Title:

Predicting the Level of Co-Activation of One Muscle Head from the Other Muscle Head of the Biceps Brachii Muscle by Linear Regression and Shallow Feedforward Neural Networks
Authors:

Nils Grimmelsmann, Malte Mechtenberg, Markus Vieth, Alexander Schulz, Barbara Hammer and Axel Schneider
Abstract: One of the challenges in close-to-body robotics is the intuitive control of exoskeletal devices which requires lag-free responses of its actuated joints. A frequently used signal domain to satisfy the required control properties is surface electromyography (sEMG). By using a Hill-type model of the muscle mainly responsible for the movement of a biological joint, which is excited by the corresponding sEMG of this muscle, the joint movement can be pre-calculated. If the muscle internal delays are used, this information can be used for an intuitive and lag-free control. So far, biomechanical limb and joint models including Hill-type muscle submodel were used. In current studies, state-of-the-art machine learning models are evaluated for this problem. Both types, classical and machine learning models, depend on the measured sEMG signals of all muscle heads of a relevant muscle and on their respective signal quality. This work introduces a method to train a virtual sEMG-sensor as a replacement for the real sEMG signal of a muscle head, thus reducing the number of real sensor electrodes on a given muscle. The virtual sensor is trained based on data from the remaining sensor. This method allows to compare the measured sEMG signal with the virtual sensor output to assess the measured signal. Furthermore, this study explains the training process and evaluates the use of the virtual sensor in a biomechanical limb model. .
Download
Paper Nr: 102
Title:

A Word Recognition Paradigm Through EEG Analysis: Imagined Speech Classification
Authors:

Francesco Iacomi, Andrea Farabbi, Maximiliano Mollura, Edoardo M. Polo, Riccardo Barbieri and Luca Mainardi
Abstract: This study presents an innovative approach for decoding imagined speech using EEG signals. The proposed analysis aims at revealing phonetic and semantic properties of imagined words through brain activity. The experimental protocol involves presenting words to subjects while recording EEG signals via a 64-channels cap. Each word is associated with three specific properties: length, presence of doubles, and category of meaning. The protocol includes fixation, cue presentation, thinking, and rest phases. EEG signals undergo meticulous preprocessing stage to mitigate noise and artifacts. Features are extracted from the processed signal, including statistical, spectral, and fractal domain measures. The dimensionality of features is reduced through statistical means. Several classifiers, (e.g., MLP, KNN, LDA, QDA), are trained and evaluated to predict mentioned properties of imagined words. An ensemble model (LDA) comprising the best 3 models mentioned above is then employed to enhance classification accuracy. Results illustrate the effectiveness of decoding imagined word properties with average accuracies of 35.2% for ”Category”, 57.2% for ”Doubles”, and 55.8% for ”Length”. By aggregating all predictions we are able to decode each single word with a mean accuracy of 11.8% (random accuracy = 8.33%) and an average word distance of 1.54. Post-classification studies on the most relevant variables and on the most discriminating channels further deepen our understanding of the proposed Imagined Speech cognitive process, showing different brain activations for different linguistic aspects.
Download
Paper Nr: 107
Title:

A New Algorithm for Innervation Zone Estimation Using Surface Electromyography: A Simulation Study Based on a Simulator for Continuous sEMGs
Authors:

Malte Mechtenberg, Nils Grimmelsmann and Axel Schneider
Abstract: In this work, a novel algorithm for the estimation of the innervation zone location within a muscle head is presented. The algorithm is able to identify innervation zone clusters within continuous surface electromyography (sEMG) recordings based on linear electrode arrays. The presented algorithm is tested in a simulation environment, which is capable of simulating EMG signals based on a common drive signal (activation). The simulator was used to generate sEMGs of six virtual muscle based on six different configurations for the respective muscle fibre distributions. The virtual muscles were each activated with a trapezoidal signal (common drive). The new algorithm was able to identify the location of the innervation zone centers with a mean absolute error of 3.8% of the inter electrode distance. In the best case, the absolute error was below 1% of the inter electrode distance.
Download
Paper Nr: 119
Title:

Gait Parameter Estimation from a Single Privacy Preserving Depth Sensor
Authors:

Yale Hartmann, Jonah Klöckner, Lucas Deichsel, Rinu E. Paul and Tanja Schultz
Abstract: Recovery from a hard fall is more difficult with age, and early detection of increased fall risk can support early prevention training. The ETAP project focuses on detecting early signifiers like step length in real-time and unobtrusively in older adult life with a single privacy-preserving depth sensor. This paper highlights our efforts to estimate a healthy individual’s skeleton and stride length and outlines how this will be transferred to care facilities. The best ResNet50-based model achieved a mean precision error of 17.49 cm per skeletal joint and stride length error of 5.73 cm on the mean stride length over 727 steps and 7.52cm over 16.67 seconds. Furthermore, 80% accuracy in step classification was achieved. These results show that gait parameter estimation is accurately possible. In the future, we aim to improve these results and build an online system with our care facility partners, transferring these findings to everyday life.
Download
Paper Nr: 144
Title:

Mapping Seismocardiogram Characteristics to Hemorrhage Status and Vascular Pressure: A Novel Approach for Triage Assessment
Authors:

Zeynep D. Gundogan and Beren Semiz
Abstract: When a mass incident occurs, determining the severity of injuries and arranging the hospital triage are of great importance to increase the survival rates. This study aims to develop a seismocardiogram (SCG)-based triage assessment system by (i) distinguishing between different levels of exsanguination, and (ii) estimating the vascular pressure values recorded from various body locations for prioritizing the triage processes and monitoring vital parameters. In this project, publicly available Wearable and Catheter-based Cardiovascular Signals During Progressive Exsanguination in a Porcine Model of Hemorrhage dataset, which includes cardiovascular signals acquired through a catheter-based system and wearable sensors during progressive exsanguination, was used. First, temporal and spectral features were extracted from the SCG signals taken at different blood-loss levels from six Yorkshire swines. Hemorrhage severity assessment was then performed through multi-class classification leveraging one vs. all approach. As the second step, four different regression models were trained for each of the right atria, aortic root, femoral artery and pulmonary capillary locations to estimate the corresponding vascular pressure values. For hemorrhage severity assessment, the accuracy, sensitivity, precision and f1-score values were all calculated to be 0.96 for the best performing model (XGBoost). For the vascular pressure estimation, (mean-absolute-error and R 2 ) pairs were calculated to be (1.54, 0.94), (2.76, 0.58), (1.29, 0.87) and (0.95, 0.90) for aortic root, femoral artery, right atrium and pulmonary capillary models, respectively. Overall, this study introduced new use areas for the SCG signal, which can potentially be utilized in the development of continuous and non-invasive monitoring systems to prioritize the triage processes and track vital parameters.
Download
Paper Nr: 152
Title:

An Insight Into Neurodegeneration: Harnessing Functional MRI Connectivity in the Diagnosis of Mild Cognitive Impairment
Authors:

Shuning Han, Zhe Sun, Kanhao Zhao, Feng Duan, Cesar F. Caiafa, Yu Zhang and Jordi Solé-Casals
Abstract: Alzheimer’s disease is a progressive form of memory loss that worsens over time. Detecting it early, when memory issues are mild, is crucial for effective interventions. Recent advancements in computer technology, specifically Graph Convolutional Networks (GCNs), have proven to be powerful tools for analyzing Magnetic Resonance Imaging (MRI) data comprehensively. In this study, we developed a GCN framework for diagnosing mild cognitive impairment (MCI) by examining the functional connectivity (FC) derived from resting-state functional MRI (rfMRI) data. Our research systematically explored various types and processing methods of FC, evaluating their performance on the OASIS-3 dataset. The experimental results revealed several key findings. On the one hand, the proposed GCN exhibited significantly superior performance over both the baseline GCN and the Support Vector Machine (SVM) models, with statistically significant differences. It attained the highest average accuracy of 80.3% and a peak accuracy of 88.2%. On the other hand, the GCN framework obtained using individual FCs showed overall slightly better performance than the one using global FCs. However, it is important to note that GCNs using global networks with appropriate connectivity can achieve comparable or even better performance than individual networks in certain cases. Finally, our results also indicate that the connectivity within specific brain regions, such as VIS, DMN, SMN, VAN, and FPC, may play a more significant role in GCN-based MRI classification for MCI diagnosis. These findings significantly contribute to the understanding of neurodegenerative disorders and offer valuable insights into the diverse applications of GCNs in brain analysis and disease detection.
Download
Paper Nr: 180
Title:

Assessing Emotion-Induced Variations of Event-Related Potentials and Heart Rate During Affective Picture Processing
Authors:

Stefania Coelli, Pierluigi Reali and Anna M. Bianchi
Abstract: Emotions are psychological responses to stimuli that can induce measurable variations in physiological parameters. While actual emotions span a continuum spectrum, they can be grouped into a finite number of classes or modeled in terms of independent dimensions, the most common of which are arousal (low to high) and valence (positive, neutral, and negative). In this work, we investigated the modulation of physiological parameters related to both the central (CNS) and the autonomic (ANS) nervous systems induced by passive and sustained affective stimulation. Specifically, an Event-Related Potential (ERP) analysis was conducted to explore the effect of the arousal and valence dimensions on cortical activation. Meanwhile, their influence on the ANS activity was evaluated through time-domain heart rate (HR) parameters. When high arousal stimuli are delivered, the experiment revealed that specific ERP components (i.e., P300 and the late positive potential, LPP) are modulated by the valence dimension, with positive and negative images inducing a stronger response than neutral stimuli. Instead, the early posterior negativity (EPN) was found to be influenced by the stimulus arousal but not by the valence of the processed pictures. Finally, HR parameters were principally modulated by the valence of the stimulation, in line with the observed ERP changes and expectations from the literature.
Download
Paper Nr: 245
Title:

Comfort Assessment Method of EEG-Based Exoskeleton Walking-Assistive Device
Authors:

Heyuan Wang, Kaitai Li, Hui Liu, Xuesong Ye and Congcong Zhou
Abstract: The study of wearable exoskeleton robotics has garnered significant attention, amidst a rapidly expanding corpus of scholarly work aimed at the empirical evaluation of the performance characteristics of robotic exoskeletons. However, quantifying comfort performance is still a significant and challenging task. This study aimed to perform comfort assessment based on EEG (Electroencephalography) signals and classical machine learning models as well as deep learning model. It involved collecting EEG data from users wearing lower limb exoskeleton walking-assistive devices for comfort assessment during walking experiments. The subjective evaluation labels of comfort were obtained using a semantic differential scale, providing comfort labels data for each participant in each trial. This study conducted a comparative analysis of three classical ML (Machine Learning) models, Naive Bayes, K-Nearest Neighbors, and Support Vector Machine models, with DL (Deep Learning) model, LSTM (Long Short-Term Memory), in terms of their accuracy for comfort assessment. The results of the analysis showed that the deep learning model, LSTM, outperformed the classical machine learning models, in terms of accuracy for evaluating comfort. Specifically, we get an accuracy of 0.91±0.12 on the LSTM model. The LSTM model demonstrated higher accuracy and better performance in capturing complex patterns and relationships within the EEG data, leading to the potential of more accurate predictions of comfort levels.
Download
Paper Nr: 275
Title:

Associating Endpoint Accuracy and Similarity of Muscle Synergies
Authors:

Chuanyun Ouyang, Liming Cai, Shuhao Yan, Tianxiang Zhang, Jun Zhu, Li Chen and Hui Liu
Abstract: Recently, Extracting the muscle synergy from surface electromyographic (sEMG) signals has become a standard method for evaluating motor control strategies during exercise. The synergy of the upper extremity in various stretch and reach tasks has been described in many studies, but few of them have analyzed the relationship between task performance and muscle synergy. This study provides an experimental device and analysis method for muscle coordination in the joystick task for the specific action of the pilots’ joystick manipulation. Eight healthy subjects performed the joystick manipulation. For upper limbs, the task content included isotonic tasks with three load levels and recorded ten muscles’ EMG and acceleration information. The muscle synergy effect was extracted and the correlation between muscle synergy similarity and manipulation performance and interaction load was studied. The experiment data showed that the manipulation performance varied under different loading conditions, but did not show significant changes in synergistic muscle structure. We found significant correlations between the similarity of some synergistic muscle structures and manipulation performance. However, between single-action performance and the average of their likeness, there was no strong correlation. Through the analysis of muscle synergy, we can determine that there is a fixed muscle synergy pattern during rocker manipulation, of which the structure is independent of the rocker load level, and muscle synergy similarity was negatively correlated with manipulation performance. The results of this study help improve the ergonomics of the flight stick and propose targeted muscle training methods to improve the precision of flight maneuvering.
Download
Paper Nr: 294
Title:

Integrated Driver Pose Estimation for Autonomous Driving
Authors:

Xiao Cao, Wei Hu and Hui Liu
Abstract: Human-machine interaction, especially driver posture estimation is important to the development of autonomous driving, which can facilitate safe and smooth driving behaviours. Besides, it also contributes to ergonomics research and human-machine interaction design for automated vehicles. The existing studies have got great achievements in body estimation, hand pose estimation, and even face feature estimation thanks to the rapid development of deep learning approaches and the upgrade of hardware equipment. However, most existing models can only process body estimation or hand estimation separately, which will impede the research on driver-vehicle interaction in autonomous driving. This is because the driving process is highly dependent on the cooperation between the body and hands behaviours. In this study, five popular deep learning models, including Simple Faster R-CNN, RootNet, PoseNet, Yolo v3, and graph convolutional neural network, are combined through a cascade method to develop an integrated model which can estimate body and hand simultaneously during the driving process. The coordinate transform system is proposed to connect models in series. Experiment results demonstrate the proposed method can produce 2D and 3D reorganization of the human body and hands simultaneously with acceptable accuracy.
Download
Short Papers
Paper Nr: 33
Title:

Preliminary Results on the Evaluation of Different Feedback Methods for the Operation of a Muscle-Controlled Serious Game
Authors:

Julia Habenicht and Elsa A. Kirchner
Abstract: Muscle-controlled serious games can improve the ability of a targeted muscle control. This aspect is important for controlling muscle-controlled prostheses or for (re-)learning motor movements. Although there are more options, all muscle-controlled serious games are using visual feedback for providing information of the current muscle activity. The aim of this study is to compare the feedback methods visual, auditory and haptic feedback for motor learning with a muscle-controlled serious game. Due to the current status of the study, in this paper only the results of visual and auditory feedback will be analysed. Subjects were divided into two groups – visual or auditory feedback. A muscle-controlled serious game was played on three days in a row by three subjects in each group. For the visual group the game provided only visual and for the auditory group it provided only auditory feedback. At the end of each session one set without any feedback was played to control the learning status. Preliminary results show a slightly better performance of the auditory group. As the results aren’t significant, more subjects are needed to get further information about the most promising feedback method for motor learning with muscle-controlled serious games.
Download
Paper Nr: 84
Title:

Evaluation of Gel and Dry Electrodes for EEG Measurement to Compare Their Suitability for Multimodal Workload Detection in Humans
Authors:

Judith Bütefür, Mathias Trampler and Elsa A. Kirchner
Abstract: In this paper we aim to investigate whether the use of dry electrodes to detect multimodal workload could be a viable way forward in the future. Therefore, we did a comparative study with gel (6 subjects) and dry electrodes (2 subjects) and analysed the data using the Task Load Index (TLI) and the power spectrum of different frequency bands. The results show that the TLI is significantly increasing for higher workload condition (p < 0.04) and expected changes in the frequency bands are significant for both gel and dry electrodes in subject-specific frequency bands. In conclusion, the results look promising, and it is worthwhile to conduct another study with more subjects using dry electrodes.
Download
Paper Nr: 105
Title:

Investigation of Artifact Contamination Impact on EEG Oscillations Towards Enhanced Motor Function Characterization
Authors:

Mojisola G. Asogbon, Oluwarotimi W. Samuel, Farid Meziane, Guanglin Li and Yongcheng Li
Abstract: The significant advancements in electroencephalography (EEG)-driven technology have led to its widespread use in assessing stroke-related conditions. Over the years, various studies have explored the potential of EEG oscillatory patterns in neurological research, with several of them giving limited attention to the signal processing techniques employed, precluding a proper understanding of EEG oscillatory patterns under various conditions. To resolve this issue, we systematically investigated how artifacts impact EEG oscillatory rhythms associated with upper limb movement-related tasks. Thus, the EEG signals of motor tasks were acquired non-invasively from healthy subjects and processed using automated artifact-attenuation methods. Subsequently, the Mu and Beta bands in the brain’s motor cortex region were extracted through time-frequency analysis and analyzed using relevant metrics. Experimental results revealed that artifacts in EEG would substantially influence the brain activation strength and response during motor tasks. Notably, signals preprocessed with Reduction of Electroencephalographic Artifacts based on Multi Wiener Filter and Enhanced Wavelet Independent Component Analysis (RELAX_MWF_wICA) showed better brain responses and high task classification performance compared to other methods and the raw signal across motor tasks. This study’s findings revealed that the choice of signal processing technique is crucial, as it would influence its analysis and interpretation, thus highlighting the need for careful consideration and usage.
Download
Paper Nr: 167
Title:

Modelling Physiological Sensor Noise to Movement-Based Virtual Reality Activities
Authors:

Phil Lopes, Nuno Fachada, Micaela Fonseca, Hugo Gamboa and Claudia Quaresma
Abstract: This position paper proposes the hypothesis that physiological noise artefacts can be classified based on the type of movements performed by participants in Virtual Reality contexts. To assess this hypothesis, a detailed research plan is proposed to study the influence of movement on the quality of the captured physiological signals. This paper argues that the proposed plan can produce a valid model for classifying noisy physiological signal features, providing insights into the influence of movement on artefacts, while contributing to the development of movement-based filters and the implementation of best practices for using various associated technologies.
Download
Paper Nr: 196
Title:

Performance Comparison of Gyrocardiogram and Seismocardiogram Signals in Valvular Heart Disease Assessment
Authors:

Ecem Erin and Beren Semiz
Abstract: Cardiovascular diseases have been identified as one of the leading causes of mortality worldwide. Among these diseases, valvular heart diseases (VHDs) have a greater impact on the population. The existing methods for VHD assessment are expensive and only applicable within clinical environments. Hence, there is a need for accessible and cost-efficient systems to provide continuous VHD monitoring. As stenosis and regurgitation are characterized by the change in the blood flow patterns, it was hypothesized that angular acceleration (gyrocardiogram, GCG) could capture the differences in blood flow and changes in cardiovascular parameters better than linear acceleration (seismocardiogram, SCG). In this work, a publicly available dataset including 36 patients with stenosis and 44 patients with regurgitation was used. The SCG and GCG signals were first divided into 10- second long segments. From each segment, five features were extracted from all axes and used to train the SCG- and GCG-based XGBoost models. Overall, the GCG-based model resulted in better performance in distinguishing between the stenosis and regurgitation cases: the precision, recall and accuracy values were 94.7, 94.5, and 94.5 for the SCG, and 96.0, 95.9 and 95.9 for the GCG, respectively. Predictive performances of SCG and GCG models on the cardiovascular parameters were also investigated and resulted in (SCG, GCG) mean absolute percent errors of (19.4, 20.6), (15.5, 14.5), (12.0, 13.1) for ejection fraction, left ventricular end diastolic dimension and left ventricle posterior wall thickness, respectively. These results showed that in addition to SCG, GCG could also be used for VHD evaluation and potentially be employed in continuous monitoring systems.
Download
Paper Nr: 197
Title:

A Hierarchical Framework for Apnea Detection and Respiration Pace Assessment Using Seismocardiogram Signals
Authors:

Berke Kizir and Beren Semiz
Abstract: Sleep constitutes one-third of human life and plays a critical role in physical repair, mental functioning, and memory consolidation. Although polysomnography (PSG) has been used to assess sleep performance; this test requires participants to visit a sleep clinic and have multiple sensors attached to their bodies. Hence, there is a need for alternative methods which can provide sleep monitoring outside clinical settings, but with clinical standards. In this work, a novel hierarchical framework was built to leverage the seismocardiogram (SCG) signals in apnea detection and respiration pace assessment using a simulated data collection protocol. In the first step of the framework, a binary Light Gradient-Boosting Machine (LGBM) model was trained to detect the breath-holding (apnea) episodes. If the prediction was not a breath-holding state, the data was fed into a multi-class LGBM model to distinguish between normal, slow and fast breathing episodes. Overall, the binary LGBM resulted in an accuracy, recall, precision and f1-score of 0.99, 0.95, 0.87 and 0.91, respectively; whereas for the multi-class case all metrics were 0.96. Additionally, the optimum window length to achieve real-time detection was determined as 5 seconds. The results show that the SCG signals hold substantial information regarding the changes in breathing patterns, thus could potentially be leveraged in the design of wearable systems as an alternative to the PSG test.
Download
Paper Nr: 218
Title:

Neuromotor Pattern of the Upper Limb in Hygiene Activities Using Electromyography and Accelerometery Technology
Authors:

Patrícia Santos, Inês Garcia, Carla Quintão and Claúdia Quaresma
Abstract: The technology is a valuable resource for movement analysis, especially for complex movement patterns such as those of the upper limb during activities of daily living (ADLs). Characterizing these patterns in healthy individuals is crucial to detect abnormal and compensatory movements resulting from neurological dysfunctions. This study aimed to characterize the neuromuscular activation pattern of the upper limb during the washing of the contralateral limb in 36 healthy individuals. The Biosignalsplux® equipment was used to monitor the activity of the main shoulder muscles, that is, Pectoralis Major (PM), Anterior Deltoid (AD), Middle Deltoid (MD), Posterior Deltoid (PD), Upper Trapezius (UT) and Lower Trapezius (LT), through electromyography (EMG) and accelerometry (ACC). The results show variations in the contraction pattern in the different phases of the activity. With this study it was possible to establish the normalized pattern of the activity of EMG and ACC of the shoulder complex and respective movement phases.
Download
Paper Nr: 226
Title:

Cramer-Rao Bound for Dipole Source Localization in Infants Using Realistic Geometry
Authors:

Aleksandar Jeremic, D. Nikolic, G. Djuricic, N. Milcanovic and Z. Jokovic
Abstract: Source localization of electrical activity in newborn infants is important from two standpoints. From an academic standpoint such insights can enable better understanding of brain development and from clinical standpoint localization of electrical activity can identify regions of the brain with higher than usual activity and possibly improve possible treatment outcomes. The electrical activity and the corresponding electroencephalography (EEG) measurements are dependant on electrical properties of brain and skull tissue i.e. corresponding conductivities and geometry. In this paper we investigate effects of realistic geometry in newborn infants by accounting for soft spots (fontanels) that are present in newborn infants. These structures have larger conductivity than regular bone tissue and hence the estimation accuracy can potentially be improved by optimally positioning EEG sensors on the surface of the skull. We generate forward model using realistic geometry and finite-element model generated by COMSOL. We utilize simplified source model consisting of single dipole source and calculate corresponding Cramer-Rao bound as a function of source intensity and locations.
Download
Paper Nr: 258
Title:

Improved PID Control Based on Temperature Compensation for the Incubation Plate of Chemiluminescent Immunoassay Analyzer
Authors:

Zhaoyang Wang, Jing Wang, Bo Liang, Xuesong Ye and Congcong Zhou
Abstract: This paper proposes an improved PID control based on temperature compensation strategy, which can dynamically adjust the target temperature value of the PID controller according to the ambient temperature and the preset temperature compensation curve, thus basically eliminating the influence of ambient temperature on the reaction liquid temperature, and ultimately achieving stable reaction liquid at the preset temperature under different ambient temperature conditions. Through experiments, it was found that after adding temperature compensation strategy to the PID control, the maximum steady-state temperature difference of the reaction liquid decreased from 0.31℃ to 0.11℃, and the coefficient of variation (CV) of Relative Light Units (RLU) decreased from 4.22% to 1.43%.
Download
Paper Nr: 265
Title:

Estimating Skull Thickness of Neonates Using Magnetic Resonance Imaging
Authors:

Aleksandar Jeremic, D. Nikolic, G. Djuricic, N. Milcanovic and Z. Jokovic
Abstract: Successful imaging of electrical activity in newborn infants is highly dependent on accurate and/or adequate representation of head representation from structural point of view. Namely, the electrical activity and the corresponding electroencephalography (EEG) measurements are dependant on electrical properties of brain and skull tissue i.e. corresponding conductivities and geometry of the skull and brain. Automated procedure for geometry/structural analysis are sparse even for adults and almost non-existent for neonates and newborn infants. In this paper we propose to develop automatic procedures for analyzing skull geometry and potentially other shapes/sizes that are relevant for electrical imaging of the cortex activity. To this purpose we propose to estimate the thickness of the skull using magnetic resonance (MR) images as a preliminary step in obtaining/estimating relevant structural parameters. Since the number of MR images is rather limited due to the age of the patients we develop a semi-supervised machine learning algorithm in which certain number of MR slices is used for training. We demonstrate applicability of our preliminary results using real MR images obtained from the University Children’s Hospital, University of Belgrade, Serbia.
Download
Paper Nr: 268
Title:

Really Can't Hold On Anymore? Physiological Indicators Versus Self-Reported Motivation Drop During Jogging
Authors:

Shiyao Zhang, Sergei Kolensnikov, Till Rennspieß, Robert Porzel, Tanja Schultz and Hui Liu
Abstract: Motivational dynamics in jogging constitute a pivotal factor influencing a runner’s performance, persistence, and overall engagement in the running activity. The manifestation of diminished motivation is concomitant with a cascade of physiological responses, capable of being represented through biological signals, for which biosignal monitoring, a common practice in evaluating athletic performance, emerges as a valuable tool. Should biosignals, as dynamic indicators during exercise, exhibit discernible shifts correlating with changes in motivation, the prospect of actively modulating motivation levels and intervening in athletes’ performance during exercise becomes feasible. This study consists of collecting comprehensive biological data, including electrocardiogram (ECG), surface electromyogram (sEMG), and respiration signals (RSP), from runners who participated in a 20-minute running session. Participants were asked to self-report a decrease in motivation during jogging. Using heart rate variability analysis, self-similarity matrix and deep learning methodologies, this work seeks to explore whether the discomforts reported and triggered by decreased motivation had discernible effects on monitored physiological signals, thus advancing our understanding of the nuanced relationship between physiological responses and motivational states in running.
Download
Paper Nr: 26
Title:

Characterization of sEMG Spectral Properties During Lower Limb Muscle Activation
Authors:

Costa-Garcia Alvaro and Shimoda Shingo
Abstract: The analysis of biological data is an effective way to extract implicit information about the human physiological condition, representing the performance of daily tasks. The use of this information as feedback for robotic systems can contribute to a smoother transition into societies with a higher level of human-robot collaboration. Superficial electromyography (sEMG) could be a powerful ally in this field, as muscle activity serves as a window into our neural system and can be measured non-invasively with relative ease. In this work, our objective is to extract spectral features that enable the classification between isometric and isotonic muscle contractions. The switching between these types of contractions during human motion has been widely linked to various physical conditions, such as muscle pain, fall prediction, postural imbalances, and stress. To achieve this goal, we recorded muscle activity during both isometric and isotonic contractions under various conditions. We conducted a time-frequency analysis on the data collected from five lower limb muscles of four healthy subjects to extract significantly relevant features containing the necessary information to discriminate between these two types of muscle activations. Our results suggest that this discrimination can be achieved through the analysis of two spectral features: the median frequency and the power contained in the frequency range between 11 and 32 Hz. Furthermore, the inclusion of the peak frequency as a third feature also enables the detection of low-frequency motion artifacts.
Download
Paper Nr: 31
Title:

Centrality of the Fingerprint Core Location
Authors:

Laurenz Ruzicka, Bernhard Strobl, Bernhard Kohn and Clemens Heitzinger
Abstract: Fingerprints have long been recognized as a unique and reliable means of personal identification. Central to the analysis and enhancement of fingerprints is the concept of the fingerprint core. Although the location of the core is used in many applications, to the best of our knowledge, this study is the first to investigate the empirical distribution of the core over a large, combined dataset of rolled, as well as plain fingerprint recordings. We identify and investigate the extent of incomplete rolling during the rolled fingerprint acquisition and investigate the centrality of the core. After correcting for the incomplete rolling, we find that the core deviates from the fingerprint center by 5.7% ± 5.2% to 7.6% ± 6.9%, depending on the finger. Additionally, we find that the assumption of normal distribution of the core position of plain fingerprint recordings cannot be rejected, but for rolled ones it can. We find the non-central Fischer distribution best describes the cores’ horizontal positions. Finally, we investigate the correlation between mean core position offset and the NFIQ 2 score and find a weak preference of the NFIQ 2 towards rolled recordings with a lower than central core.
Download
Paper Nr: 69
Title:

Feature Selection Improves Speech Based Parkinson's Disease Detection Performance
Authors:

Ayşe N. Tekindor and Eda A. Aydın
Abstract: Parkinson’s disease (PD) is a neurodegenerative disorder that is caused by decrease in dopamine levels in the brain. There is currently no cure for PD; however, the progression of the disease can be brought under control by diagnosis made in early stages. Studies have shown that speech impairments are early symptoms of PD. In this study, an approach for the early diagnosis of patients with PD using speech based features was proposed. In order to detect the PD, four feature groups such as Bark Spectrum coefficients, Mel Frequency Cepstral Coefficients (MFCCs), Gammatone Cepstral Coefficients (GTCCs), and Spectral-Temporal Features were created. Minimum Redundancy Maximum Relevance (mRMR) based feature selection was applied to each feature group. Three classifiers including decision tree, Naive Bayes, and support vector machine were employed to evaluate the performance of the feature sets. The proposed method was validated on the Italian speech dataset. Feature selection improved the PD diagnosing performance, especially for the Naive Bayes model which obtained 96.01% accuracy by overall feature selection and 96.17% by group-based feature selection.
Download
Paper Nr: 74
Title:

Hand Movement Recognition Based on Fusion of Myography Signals
Authors:

Shili W. Eddine, Youssef Serrestou, Slim Yacoub, Ali H. Al-Timemy and Kosai Raoof
Abstract: This article presents a hand movement classification system that combines acoustic myography (AMG) signals, electromyography (EMG) signals and mechanomyogram signal (MMG) data. The system aims to accurately predict hand movements, with the potential to improve the control of hand prostheses. A dataset was collected from 9 individuals who repeated 10 times each of 4 hand movements (hand close, hand open, fine pinch and index flexion). The system, with a Support Vector Machine (SVM) classifier, achieved an accuracy score of 97%, demonstrating its potential for real-time hand prosthesis control. The combination of AMG, EMG, and MMG signals proved to be effective in accurately classifying hand movements.
Download
Paper Nr: 81
Title:

Prediction of Oxygen Saturation from Graphene Respiratory Signals with PPG Trained DNN
Authors:

Bojana Koteska, Ana M. Bogdanova, Teodora Vićentić, Stefan D. Ilić, Miona Tomić and Marko Spasenović
Abstract: This paper explores the feasibility of using wearable laser-induced graphene (LIG) sensors to estimate oxygen saturation (SpO2) as an alternative to traditional photoplethysmography (PPG) oximeters, particularly in mass casualty triage scenarios. Positioned on the chest, the LIG sensor continuously monitors respiratory signals in real-time. The study leverages deep neural network (DNN) trained on PPG signals to process LIG respiratory signals, revealing promising results. Key performance metrics include a mean squared error (MSE) of 0.152, a mean absolute error (MAE) of 1.13, a root mean square error (RMSE) of 1.23, and an R2 score of 0.68. This innovative approach, combining PPG and respiratory signals from graphene, offers a potential solution for 2D sensors in emergency situations, enhancing the monitoring and management of various medical conditions. However, further investigation is required to establish the clinical applications and correlations between these signals. This study marks a significant step toward advancing wearable sensor technology for critical healthcare scenarios.
Download
Paper Nr: 112
Title:

ResNet-101 Empowered Deep Learning for Breast Cancer Ultrasound Image Classification
Authors:

Agnesh C. Yadav, Maheshkumar H. Kolekar and Mukesh K. Zope
Abstract: In the modern era, accurate breast cancer classification plays a crucial role in early detection and treatment planning. This article introduces a modified ResNet-101 architecture tailored specifically for classifying breast cancer using ultrasound images. The ultrasound images undergo pre-processing before passing through our adapted ResNet-101 model, which includes the integration of shortcut connections to enhance gradient stability and deep structure adaptability for effective learning and classification. The dataset comprises 780 images categorized into normal, benign, and malignant cases. To address class imbalance, data augmentation techniques are employed, enriching diversity and enhancing modeling precision. The proposed model achieves exceptional performance, boasting precision, recall, F1-score, and accuracy values of 0.9855, 0.9677, 0.9756, and 0.9743, respectively. The comparative analysis highlights the superiority of our model over existing techniques. Furthermore, we explore its potential for clinical application using real-world datasets. Our findings indicate significant promise in revolutionizing breast cancer detection, offering a robust tool for early and accurate diagnosis with the potential to impact patient outcomes greatly.
Download
Paper Nr: 117
Title:

Wavelet Based Feature Extraction for Multi-Model Ensemble Approach for Mental Workload Classification Using EEG
Authors:

Fiza Parveen and Arnav Bhavsar
Abstract: Mental workload is a crucial aspect of cognitive processing as it reflects how much of our working memory is engaged. Studying n-back tasks of varying complexity, has been a popular way to explore the relationship between mental workload and EEG patterns. However there is still scope of improvement in achieving good performance in such a mapping. In this work, we address the classification of EEG patterns corresponding to different n-back tasks. We use publicly available n-back dataset, comprising 0-back, 2-back, and 3-back tasks to represent low, medium, and high levels of mental workload, respectively. We use wavelet-based signal decomposition technique to compute multi-resolution representation having both time and frequency patterns. This is followed by extracting a variety of hand crafted feature. We train different XGBoost models for two level and three level mental workload classification. Furthermore, we employ ensemble techniques at different levels to better categorize EEG signals. Our approach also involves finding channels that are most significant for classification of highly complex 2-back and 3-back task EEG data.
Download