Biomedical Signal Processing and Control

Considering the properties of the empirical mode decomposition to extract from a signal its natural oscillatory components known as intrinsic mode functions (IMFs), the spectral analysis of these IMFs could provide a novel alternative for the quantitative EEG analysis without a priori establish more or less arbitrary band limits. This approach has begun to be used in the last years for studies of EEG records of patients included in database repositories or including a low number of individuals or of limited EEG leads, but a detailed study in healthy humans has not yet been reported. Therefore, in this study the aims were to explore and describe the main spectral indices of the IMFs of the EEG in healthy humans using a method based on the FFT and another on the Hilbert-Huang transform (HHT). The EEG of 34 healthy volunteers was recorded and decomposed using a recently developed multivariate empirical mode decomposition algorithm. Extracted IMFs were submitted to spectral analysis with, and the results were compared with an ANOVA test. The first six decomposed IMFs from the EEG showed frequency values in the range of the classical bands of the EEG (1.5 to 56 Hz). Both methods showed in general similar results for mean weighted frequencies and estimations of power spectral density, although the HHT is recommended because of its better frequency resolution. It was shown the presence of the mode-mixing problem producing a slight overlapping of spectral frequencies mainly between the IMF3 and IMF4 modes.

Cardiovascular disease is the main cause of death worldwide. An electrocardiogram (ECG) signals is typically used as the first diagnosis tool to detect abnormality in the heart signal. Reliable detection of R-peak in the ECG signal indicates various heart malfunctions (e.g., arrhythmia) and allows for proactive prevention of death due to cardiovascular disease. Though several R-peak detection methods have been proposed, the existence of noise in ECG signals and changes in QRS morphology compromise the robustness and reliability of these methods. Therefore, the need for a reliable detection of R-peak is crucial for diagnosis of heart abnormalities. This paper introduces a time-efficient and novel continuous wavelet transform (CWT) based method for R-peak detection. The proposed method first transforms the ECG signal in to time-frequency spectrum. The contributions of different frequencies at every time point are then calculated from the time-frequency spectrum to efficiently reduce the impact of noise and generate a summed frequency signal. A threshold technique is also proposed to detect the R-peak from the newly generated signal allows. The MIT-BIH arrhythmia database is used as a reference for validation and comparison of the proposed method with the results of other existing R-peak detection methods. The experimental results prove the efficiency and robustness of the developed method on noisy ECG signals with changes in QRS morphology with 99.87% sensitivity, 99.76% positive prediction value and a detection error rate of only 0.37%. In addition to the high accuracy in detecting R-peaks, the ease-to-use and fast-processing make the proposed method an efficient and reliable tool for real-time abnormality detection in ECG signal.

The analysis of EEG microstates is a useful method for exploring large-scale networks and brain dynamics. In addition to the often-reported microstates, or so-called "canonical microstates", another topography has been reported in the literature - topography with a prominent straight line separating positive and negative values that extends from the nasion to the inion (vertical topography - VT). This topography was also revealed in our simultaneous EEG/fMRI and shielded cabin EEG data collected from 77 participants. Following analyses based on human and phantom data, we conclude that VT partially reflects artifacts caused by unspecified movements of the EEG cap and its metallic components. Our conclusion is supported by evaluation of spatiotemporal characteristics of VT estimated from EEG acquired under various conditions, especially by significant correlation between the framewise displacement (obtained from human EEG/fMRI) and the temporal characteristics of VT. We recommend cautious interpretation of VT when revealed in the data. Its very presence as a resulting topography may affect the spatiotemporal parameters of the other microstates and distorts the shapes of the other topographies. Key pointsO_LIVertical microstate topography (VT) is often present in EEG/fMRI data. C_LIO_LIIn EEG/fMRI, VT mainly represents artifacts. C_LIO_LIAnalysis of EEG microstates is disrupted by the presence of VT. C_LI

The complexity of biological signals has been proposed to reflect the adaptability of a given biological system to different environments. Two measures of complexity--multiscale entropy (MSE) and entropy of entropy (EoE)--have been proposed, to evaluate the complexity of heart rate signals from different perspectives. The MSE evaluates the information content of a long time series across multiple temporal scales, while the EoE characterizes variation in amount of information, which is interpreted as the \"state changing,\" of segments in a time series. However, both are problematic when analyzing white noise and are sensitive to data size. Therefore, based on the concept of \"state changing,\" we propose state change probability (SCP) as a measure of complexity. SCP utilizes a statistical hypothesis test to determine the physiological state changes between two consecutive segments in heart rate signals. The SCP value is defined as the ratio of the number of state changes to total number of consecutive segment pairs. Two common statistical tests, the t-test and Wilcoxon rank-sum test, were separately used in the SCP algorithm for comparison, yielding similar results. The SCP method is capable of reasonably evaluating the complexity of white noise and other signals, including 1/f noise, periodic signals, and heart rate signals, from healthy subjects, as well as subjects with congestive heart failure or atrial fibrillation. The SCP method is also insensitive to data size. A universal SCP threshold value can be applied, to differentiate between healthy and pathological subjects for data sizes ranging from 100 to 10,000 points. The SCP algorithm is slightly better than the EoE method when differentiating between subjects, and is superior to the MSE method.

This paper aims to apply machine learning techniques to an automated epileptic seizure detection using EEG signals to help neurologists in a time-consuming diagnostic process. We employ two approaches based on convolution neural networks (CNNs) and artificial neural networks (ANNs) to provide a probability of seizure occurrence in a windowed EEG recording of 18 channels. In order to extract relevant features based on time, frequency, and time-frequency domains for these networks, we consider an improvement of the Bayesian error rate from a baseline. Features of which the improvement rates are higher than the significant level are considered. These dominant features extracted from all EEG channels are concatenated as the input for ANN with 7 hidden layers, while the input of CNN is taken as raw multi-channel EEG signals. Using multi-concept of deep CNN in image processing, we exploit 2D-filter decomposition to handle the signal in spatial and temporal domains. Our experiments based on CHB-MIT Scalp EEG Database showed that both ANN and CNN were able to perform with the overall accuracy of up to 99.07% and F1-score of up to 77.04%. ANN with dominant features is more capable of detecting seizure events than CNN whereas CNN requiring no feature extraction is slightly better than ANN in classification accuracy.

Artifact elimination has become an inseparable part while processing electroencephalogram (EEG) in most brain computer interface (BCI) applications. Scientists have tried to introduce effective and efficient methods which can remove artifacts and also reserve desire information pertaining to brain activity. Blind source separation (BSS) methods have been receiving a great deal of attention in recent decades since they are considered routine and standard signal processing tools and are commonly used to eliminate artifacts and noise. Most studies, mainly EEG-related ones, apply BSS methods in preprocessing sections to achieve better results. On the other hand, BSS methods should be followed by a classifier in order to identify artifactual sources and remove them in next steps. Therefore, artifact identification is always a challenging problem while employing BSS methods. Additionally, removing all detected artifactual components leads to loss of information since some desire information related to neural activity leaks to these sources. So, an approach should be employed to suppress the artifacts and reserve neural activity. In this study, a new hybrid method is proposed to automatically separate and identify electroencephalogram (EEG) sources with the aim of classifying and removing artifacts. Automated source identification is still a challenge. Researchers have always made efforts to propose precise, fast and automated source verification methods. Reliable source identification has always been of great importance. This paper addresses blind source separation based on second order blind identification (SOBI) as it is reportedly one of the most effective methods in EEG source separation problems. Then a new method for source verification is introduced which takes advantage of components phase spaces and their dynamics. A new state space called angle space (AS) is introduced and features are extracted based on the angle plot (AP) and Poincare planes. Identified artifactual sources are eliminated using stationary wavelet transform (SWT). Simulated, semi-simulated and real EEG signals are employed to evaluate the proposed method. Different simulations are performed and performance indices are reported. Results show that the proposed method outperforms most recent studies in this subject.

Despite a large effort in EEG signal processing, classification algorithms, deep-learning approach, predicting the sensitivity to general anesthesia (GA) remains a daunting hurdle. We compare here the ability of the Bispectral Index (BIS), developed more that twenty years ago to monitor the depth of anesthesia, with the real-time checkpoint-decomposition algorithm (CDA) to evaluate the patient sensitivity from the early induction phase of GA. Using EEG recorded in children anesthetised with propofol, we computed three parameters extracted from the BIS: 1-the minimum value (nadir) of the BIS, 2-the time to reach the minimum and 3-the duration spent below 40 during the first 10 minutes. Using a logistic regression procedure, we report that these parameters provide a poor prediction of sensitivity compared to the CDA, that combined the first occurrence time of iso-electric EEG traces, fraction of suppressions of the -band and its first occurrence time. Finally, we correlate the BIS values with the maximum power frequency of the -band, the proportion of -suppressions (S) and iso-electric suppressions (IES) as well as the and{delta} power ratios. To conclude, the checkpoint-decomposition algorithm complements the EEG indices such as the BIS to anticipate the sensitivity to GA.

Atrial Fibrillation (AFIB) is a common atrial arrhythmia that affects millions of people worldwide. However, most of the time, AFIB is paroxysmal and can pass unnoticed in medical exams therefore regular screening is required. This paper proposes machine learning methods to detect AFIB from short-term ECG and PPG signals. Several experiments were conducted across five different databases with three of them containing ECG signals and the other two consisting of only PPG signals. A total of 269,842 signal segments were analyzed across all datasets (212,266 were of normal sinus rhythm (NSR) and 57,576 corresponded to AFIB segments). Experiments were conducted to investigate the hypothesis that a machine learning model trained to predict AFIB from ECG segments, could be used to predict AFIB from PPG segments. A random forest machine learning algorithm achieved the best accuracy and achieved a 90% accuracy rate on the UMMC dataset (216 samples) and a 97% accuracy rate on the MIMIC-III dataset (2,134 samples). The ability to detect AFIB with significant accuracy using machine learning algorithms from PPG signals, which can be acquired via non-invasive contact or contactless, is a promising step forward toward the goal of achieving large-scale screening for AFIB.

This study aims to identify seizures in four different stages among epileptic patients, utilizing the Bangalore Epilepsy Dataset (BEED). This dataset, which has 16 channels, was sourced from the UCI Machine Learning Repository. Initially, the data underwent preprocessing through UMAP for dimensionality reduction. This was succeeded by feature extraction via the Fast Fourier Transform (FFT), which transformed the scaled signals into the frequency domain to capture their spectral characteristics. The findings show that a two-level ensemble model surpasses the performance of leading methods, reaching an accuracy rate of 97.06%. The models performance was confirmed through stringent nested cross-validation, guaranteeing consistency across all dataset folds. The models potential for real-time deployment on Edge and Internet of Things (IoT) devices is underscored by these findings.

AO_SCPLOWBSTRACTC_SCPLOWAssessing signal quality is crucial for photoplethysmogram analysis, yet a precise mathematical model for defining signal quality is often lacking, posing challenges in the quantitative analysis. To tackle this problem, we propose a Signal Quality Index (SQI) based on the adaptive non-harmonic model (ANHM) and a Signal Quality Assessment (SQA) model, which is trained using the boosting learning algorithm. The effectiveness of the proposed SQA model is tested on publicly available databases with experts annotations. Result: The DaLiA database [20] is used to train the SQA model, which achieves favorable accuracy and macro-F1 scores in other public databases (accuracy 0.83, 0.76 and 0.87 and macro-F1 0.81, 0.75 and 0.87 for DaLiA-testing dataset, TROIKA dataset [31], and WESAD dataset [23], respectively). This preliminary result shows that the ANHM model and the model-based SQI have potential for establishing an interpretable SQA system.

For the characterisation and diagnosis of neurological disorders, dynamical, causal and crossfrequency coupling analysis using the EEG has gained considerable attention. Due to high computational costs in implementing some of these methods, the selection of important EEG channels is crucial. The channel selection method should be able to accommodate non-linear and spatiotemporal interactions among EEG channels. In neuroscience, different measures of (dis)similarity are used to quantify functional connectivity between EEG channels. Brain regions functionally connected under one measure do not necessarily imply the same with another measure, as they could even be disconnected. Therefore, developing a generic measure of (dis)similarity is important in channel selection. In this paper, learning of spatial and temporal structures within the data is achieved by using kernel-based nonlinear manifold learning, where the positive semi-definite kernel is a generalisation of various (dis)similarity measures. We introduce a novel EEG channel selection method to determine which channel interrelationships are more important for the in-depth neural dynamical analysis, such as understanding the effect of neurodegeneration, e.g. Alzheimers disease (AD), on global and local brain dynamics. The proposed channel selection methodology uses kernel-based nonlinear manifold learning via Isomap and Gaussian Process Latent Variable Model (Isomap-GPLVM). The Isomap-GPLVM method is employed to learn the spatial and temporal local similarities and global dissimilarities present within the EEG data structures. The resulting kernel (dis)similarity matrix is used as a measure of synchrony, i.e. linear and nonlinear functional connectivity, between EEG channels. Based on this information, linear Support Vector Machine (SVM) classification with Monte-Carlo cross-validation is then used to determine the most important spatio-temporal channel inter-relationships that can well distinguish a group of patients from a cohort of age-matched healthy controls (HC). In this work, the analysis of EEG data from HC and patients with mild to moderate AD is presented as a case study. Considering all pairwise EEG channel combinations, our analysis shows that functional connectivity between bipolar channels within temporal, parietal and occipital regions can distinguish well between mild to moderate AD and HC groups. Furthermore, while only considering connectivity with respect to each EEG channel. Our results indicate that connectivity of EEG channels along the fronto-parietal with other channels are important in diagnosing mild to moderate AD.

This paper proposes a simple algorithm to remove noise and artifact from multichannel data. Data are processed trial by trial: for each trial the covariance matrix of the trial is diagonalized together with that of the full data to reveal the subspace that is - locally - most eccentric relative to other trials. That subspace is then projected out from the data of that trial. This algorithm addresses a fundamental limitation of standard linear analysis methods (e.g. ICA) that assume that brain and artifact are linearly separable within the data. That assumption fails if there are more sources, including noise and brain sources, than data channels. The algorithm captitalizes on the fact that, if enough of those sources are temporally sparse, linear separation may succeed locally in time. The paper explains the rationale, describes the algorithm, and evaluates the outcome using synthetic and real brain data.

This study tests the feasibility of estimating some time-domain heart rate variability indices (the standard deviation of the RR time series, SDNN, and the standard deviation of the differentiated RR time series, or RMSSD) from smoothed and rounded to the nearest beat per minute heart period time series using shallow neural networks. These time series are often stored in wearable devices instead of the beat-to-beat RR time series. Because the algorithm for obtaining the recorded mean heart rate in wearable devices is often not disclosed, this study test different hypothetic sampling strategies and smoothers. Sixteen features extracted from 5 minute smoothed heart period time series were employed to train, validate, and test shallow neural networks in order to provide estimates of the SDNN and RMSSD indices from freely available public databases RR time series. The results show that, using the proposed features, the median relative error (averaged for each database) in the SDNN ranges from 2% to 14% depending on the smoothness, sampling strategy, and database. The RMSSD is harder to estimate, and its median relative error ranges from 6% to 32%. The proposed methodology can be easily extended to other averaged heart rate time series, HRV indices and supervised learning algorithms

Rodent hippocampal power spectra comprise of periodic and aperiodic components. The periodic components (brain rhythms) contain information about the behavioral or cognitive state of the animal. The aperiodic components are rarely studied and their functionality is not well understood, though have shown to be correlated with animals age or the excitation-inhibition ratio of the brain region. To study these components in the mouse hippocampus we modified the existing open-source FOOOF toolbox, which was originally optimized for EEG data. First, using simulated data, we show that our modifications decrease the error in assessment of the periodic components from 3% to 0.1%. Second, using tetrode electrophysiological signals, we compare the aperiodic activity within mice hippocampal sub-regions, CA1 and dentate gyrus (DG). Our optimization of FOOOF improved the aperiodic assessment errors by about 50% and were critical in making the first-ever assessments of the aperiodic components in these brain regions. Our results show that the aperiodic parameters in the DG are multi-fold and significantly larger than in CA1, revealing higher excitation and longer timescales in CA1 than DG. Our work highlights the subtle differences in electrophysiology field potentials between hippocampal sub-regions, and presents the improvements needed in the existing open-source toolbox to be able to see such differences.

This article examines the possibility of using non-linear models(Support Vector Regression) to model the single channel EEG signals from psychiatric patients and a group of normal participants, to predict psychology trait ratings, like attention, anxiety, alertness, fatigue, sleepiness and depression. It used linear models as benchmarks, and the results showed non-linear models outperformed the benchmarks, as well as more advanced linear methods, like principle component regression. It is thus concluded that using single channel in practical situations to monitor these traits would be possible.

EEG signals of healthy individuals and epileptic patients, when treated as time series of evolving dynamical systems, are found to display characteristic differences in the behavior of the unstable periodic orbits (UPO), marking the transition from regular periodic variations to self-similar dynamics. The UPO, manifesting as broad resonances in the Fourier power spectra, are quite prominent in their presence in the normal signals and are either absent or considerably weakened with a shift towards lower frequency in the epileptic condition. The weighted average and visibility power computed for the UPO region are found to distinguish epileptic seizure from healthy individuals EEG. Remarkably, the unstable periodic motion for healthy ones is well described by damped harmonic motion, the orbits displaying smooth dynamics. In contrast, the epileptic cases show bi-stability and piecewise linear motion for the larger orbits, exhibiting large sudden jumps in the velocity (referred to the rate of change of the EEG potentials), characteristically different from the healthy cases, highlighting the efficacy of the UPO as biomarkers. For both the regions, 8-14Hz UPO and 40-45Hz resonance, we used data driven analysis to derive the system dynamics in terms of sinusoidal functions, which reveal the presence of higher harmonics, confirming nonlinearity of the underlying system and leading to quantification of the discernible differences between the healthy and epileptic patients. The gamma wave region in the 40-45Hz range, connecting the conscious and the unconscious states of the brain, reveals well-structured coherence phenomena, in addition to the prominent resonance, which potentially can be used as a biomarker for the epileptic seizure. The wavelet scalogram analysis for both UPO and 40-45Hz region also clearly differentiates the healthy condition from epileptic seizure, confirming the above dynamical picture, depicting the higher harmonic generation, and intermixing of different modes in these two regions of interest. SignificanceUnstable periodic orbits are demonstrated as faithful biomarkers for detecting seizure, being prominently present in the Fourier power spectra of the EEG signals of the healthy individuals and either being absent or significantly suppressed for the epileptic cases, showing distinctly different behavior for the unstable orbits, in the two cases. A phase space study, with EEG potential and its rate of change as coordinate and corresponding velocity, clearly delineates the dynamics in healthy and diseased individuals, demonstrating the absence or weakening of UPO, that can be a reliable bio-signature for the epileptic seizure. The phase-space analysis in the gamma region also shows specific signatures in the form of coherent oscillations and higher harmonic generation, further confirmed through wavelet analysis.

The goal of this work was to use EEG data from epilepsy patients and analyze it for connectivity between brain regions using cross-power spectral density (CPSD). We took data from 76 EEG sessions by removing those with epileptic seizures. Then, we calculated CPSD for the 210 electrode pairs for each patient. Initially, we calculated CPSD in the entire sample and then stratified it based on frequency bands related to different brain states. We observed the brain hubs, the default mode network (DMN) and the anti-correlated network attached to the DMN. The connectivity of the brain changed with change in frequency and hence, brain states. EEG analysis with CPSD is a relatively inexpensive, longer duration and more convenient method over fMRI but provides similar information. Similar study can be used to identify brain connectivity patterns while brain is performing a specific task by collecting EEG data for the task.

Emotions are constructed and emerge through the dynamic interaction of multiple components. It is difficult to capture the dynamics using static or artificial stimuli. Hence, there is a need for an experiment paradigm using ecologically valid film stimuli. The data set described in this work results from an attempt to capture felt emotional experience at a particular point in time using physiological measures like EEG, ECG and EMG as well as self-reported scales. Sixteen emotional film stimuli were used from the film stimuli dataset validated in the Indian population. Participants self-reported the felt emotional category. Both the raw and pre-processed data are provided along with the pre-processing pipeline. The paradigm we have adopted is new which we have termed as Emotional Event Marker Paradigm (EEMP). Hence, the dataset has unique information about temporal markers of emotional experiences while watching the film stimuli, which is not available with any data to date. It is the first EEG data with emotional film stimuli on the Indian population. This data can be utilized to study dynamic activation and connectivity in a whole-brain source localization study, understand the mutual interactions between the central and autonomic nervous system, understand temporal hierarchy using multi-resolution tools, and perform machine learning-based classification and complex networks analysis associated with emotions.

Brain responses related to working memory originate from distinct brain areas and oscillate at different frequencies. EEG signals with high temporal correlation can effectively capture these responses. Therefore, estimating the functional connectivity of EEG for working memory protocols in different frequency bands plays a significant role in analyzing the brain dynamics with increasing memory and cognitive loads, which remains largely unexplored. The present study introduces a Bayesian structure learning algorithm to learn the functional connectivity of EEG in sensor space. Next, the functional connectivity graphs are taken as input to the graph convolutional network to classify the working memory loads. The intrasubject (subject-specific) classification performed on 154 subjects for six different verbal working memory loads produced the highest classification accuracy of 96% and average classification accuracy of 89%, outperforming state-of-the-art classification models proposed in the literature. Furthermore, the proposed Bayesian structure learning algorithm is compared with state-of-the-art functional connectivity estimation methods through intersubject and intrasubject statistical analysis of variance. The results also show that the alpha and theta bands have better classification accuracy than the beta band.

Self-supervised learning has emerged as a promising paradigm for enhancing the analysis of physiological signals, particularly Electrocar-diogram (ECG) and Photoplethysmogram (PPG) data. This review paper surveys the application of self-supervised learning techniques in the domain of ECG and PPG signal analysis. Traditional supervised methods often rely on labeled data, which can be limited and costly to acquire in medical contexts. Self-supervised learning leverages the inherent structure and temporal dependencies within ECG and PPG signals to train models without explicit annotations. By exploiting pretext tasks such as predicting time intervals, missing samples, or temporal order, self-supervised approaches can learn meaningful repre- sentations that capture crucial information for subsequent downstream tasks. This paper provides an overview of key self-supervised methods applied to ECG and PPG data, highlighting their advantages and chal- lenges. Furthermore, it discusses the transferability of learned represen- tations to various clinical applications, including arrhythmia detection, anomaly detection, and heart rate variability analysis. Through this comprehensive review, we shed light on the potential of self-supervised learning to revolutionize ECG and PPG signal processing, ulti- mately contributing to improved healthcare diagnostics and monitoring.