Biomedical Signal Processing and Control

It is generally assumed that the analysis of intracranial pressure (ICP) waveforms could be used for the detection of multiple cerebral pathophysiologies. A main obstacle for the analysis of ICP waveforms is given by the large variation of their generating signal, the arterial blood pressure (ABP). Using extended principal component analysis (PCA) we show that it is possible to distinguish between ICP waveforms generated by pathological ABP waveforms, e.g. in the case of a heart failure, without loosing information about the state of the cerebral compliance. We also create a dataset for ICP pulse classification that can be used to train models to distinguish between an intact and diminished cerebral compliance, as a function of the ICP-generating ABP pulse. As a baseline for classification we build a fully convolutional network (FCN) that reaches high performance with relatively few data.

BackgroundThe study of cerebral activity during sleep using the electroencephalograph (EEG) is a major research field in neuroscience. Despite the rich literature in this field, the automatic and accurate categorization of wake-sleep stages remains an open problem. New MethodA robust model-based Kalman filtering scheme is proposed for tracking the poles of a second order time-varying autoregressive model fitted over the EEG acquired during different wake/sleep stages. The pole angle/phase is regarded as the dominant frequency of the EEG spectrum (known as the instantaneous frequency in literature). The frequency resolution is improved by splitting the wide frequency band to subbands corresponding to well-known brain rhythms. Using recent findings in field of EEG phase/frequency tracking, the instantaneous envelope of the narrow-band signals analytic form is also tracked as a complementary feature. ResultsThe minimal set of instantaneous frequency and envelope features is employed in three classification schemes, using training labels from R&k and AASM sleep scoring standards. The LDA classifier resulted in the highest performance using the proposed feature set. Comparison with Existing MethodsThe proposed method resulted in a higher mean decoding accuracy and a lower standard deviation on the entire dataset, as compared with state-of-the-art techniques. ConclusionsThe accurate tracking of the instantaneous frequency and envelope are highly informative for sleep stage scoring. The proposed method is shown to have additional applications, including the prediction of wake-sleep transition, which can be used for drowsiness detection from the EEG.

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.

Accurate assessment of consciousness during general anesthesia is crucial for optimizing anesthetic dosage and patient safety. Current electroencephalogram-based monitoring devices can be inaccurate or unreliable in specific surgical contexts (e.g. cephalic procedures). This study investigated the feasibility of using electrocardiogram (ECG) features and machine learning to differentiate between awake and anesthetized states. A cohort of 48 patients undergoing surgery under general anesthesia at the Tours hospital was recruited. ECG-derived features were extracted, including spectral power, heart rate variability and complexity metrics, as well as heart rate fragmentation indices (HRF). These features were augmented by a range of physiological variables. The aim was to evaluate a number of machine learning algorithms in order to identify the most appropriate method for classifying the awake and anesthetized states. The gradient boosting algorithm achieved the highest accuracy (0.84). Notably, HRF metrics exhibited the strongest predictive power across all models tested. The generalizability of this ECG-based approach was further assessed using public datasets (VitalDB, Fantasia, and MIT-BIH Polysomnographic), achieving accuracies above 0.80. This study provides evidence that ECG-based classification methods can effectively distinguish awake from anesthetized states, with HRF indices playing a pivotal role in this classification. Author summaryGeneral anesthesia monitoring is critical for optimizing patient safety and outcomes. While electroencephalogram (EEG)-based systems are commonly used, they have limitations in accuracy and applicability, particularly in cases where EEG electrodes placement is challenging or impossible, such as during cephalic surgeries or when patients have forehead skin lesions. Here, a novel approach using electrocardiogram (ECG) signals and machine learning techniques was used to differentiate between awake and anesthetized states during surgery. A total of 48 patients undergoing surgical procedures under general anaesthesia at the Tours hospital were selected for inclusion in the study. This investigation focused on heart rate fragmentation indices, metrics designed for assessing biological versus chronological age, derived from ECG recordings. The gradient boosting algorithm demonstrates performance comparable to leading methods reported in the literature for this classification task. Importantly, model generalizability was confirm through successful application to publicly available datasets. This article highlights the potential of ECG signals as an alternative source for deriving depth of anesthesia indices, offering increased versatility in clinical settings where EEG monitoring is challenging or contraindicated.

ObjectiveFor more than 25 million drug-resistant epilepsy patients, surgical intervention aiming at resecting brain regions where seizures arise is often the only alternative therapy. However, the identification of this epileptogenic zone (EZ) is often imprecise which may affect post-surgical outcomes (PSOs). Interictal high-frequency oscillations (HFOs) have been revealed to be reliable biomarkers in delineating EZ. In this paper, an analytical methodology aiming at automated detection and classification of interictal HFOs is proposed to improve the identification of EZ. Furthermore, the detected high-rate HFO areas were compared with the seizure onset zones (SOZs) and resected areas to investigate their clinical relevance in predicting PSOs.\n\nMethodsFIR band-pass filtering as well as a combination of time-series local energy, peak, and duration analysis were utilized to identify high-rate HFO areas in interictal, multi-channel intracranial electroencephalographic (iEEG) recordings. The detected HFOs were then classified into fast-ripple (FR), ripple (R), and fast-ripple concurrent with ripple (FRandR) events.\n\nResultsThe proposed method resulted in sensitivity of 91.08% and false discovery rate of 7.32%. Moreover, it was found that the detected HFO-FRandR areas in concordance with the SOZs would have better delineated the EZ for each patient, while limiting the area of the brain required to be resected.\n\nConclusionTesting on a dataset of 20 patients has supported the feasibility of using this method to provide an automated algorithm to better delineate the EZ.\n\nSignificanceThe proposed methodology may significantly improve the precision by which pathological brain tissue can be identified.

Arterial blood pressure (ABP) and photoplethysmography (PPG) waveforms contain valuable clinical information and play a crucial role in cardiovascular health monitoring, medical research, and managing medical conditions. The features extracted from PPG waveforms have various clinical applications ranging from blood pressure monitoring to nociception monitoring, while features from ABP waveforms can be used to calculate cardiac output and predict hypertension or hypotension. In recent years, many machine learning models have been proposed to utilize both PPG and ABP waveform features for these healthcare applications. However, the lack of standardized tools for extracting features from these waveforms could potentially affect their clinical effectiveness. In this paper, we propose an automatic signal processing tool for extracting features from ABP and PPG waveforms. Additionally, we generated a PPG feature library from a large perioperative dataset comprising 17,327 patients using the proposed tool. This PPG feature library can be used to explore the potential of these extracted features to develop machine learning models for non-invasive blood pressure estimation.

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.

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.

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.

In the application of motor imagery brain-computer interface system, high-density leads bring redundant noise, which leads to time-consuming system operation and poor performance. A channel selection strategy based on brain function network is proposed. This method introduces Synchronization likelihood was used as a connection index to construct a motor imagery brain functional network, and the centrality analysis of the constructed network was used to select the combination of strong motor-related leads. Experiments were carried out on the EEG datasets dataset IVa of the 3rd International Brain-Computer Interface Competition and dataset I of the 4th International Brain-Computer Interface Competition, and 27 high-quality channels were selected from the 118 channels of dataset IVa, as well as 16 high-quality channels were selected from the 59 channels of dataset I. Finally, the CSP algorithm and support vector machine are used to extract features and classify. The experimental results show that the proposed channel selection strategy can greatly reduce the number of channels while obtaining higher recognition accuracy, which verifies the practicability and effectiveness of the proposed method.

In this paper, we aim to propose a model for automatic sleep stage classification based on physiological signals acquired by Dreem Headband and extreme gradient boosting (XGBoost) method. The dataset used in this study belongs to a challenge competition, namely as "Challenge Data", held in 2017-2018, and is publicly available on their website. Recordings, includes 4 EEG channels (FpZ-O1, FpZ-O2, FpZ-F7, F8-F7), 2 Pulse oximeter (RED & infra-red), and 3 accelerometer channels (X, Y, Z). In this work, sleep stages have been scored according to the AASM standard. Different features were extracted from the physiological signals after applying a preprocessing step. Each of the elicited features from EEG and PPG signals is falling into one of the three categories: time-domain, frequency domain, or non-linear features. Moreover, ancillary features including body movement, frequency features, breathing frequency, and respiration rate variability were also extracted from the accelerometer signal. Significance of the extracted features was examined through the Kruskal Wallis test, and features with P-value>0.01 were discarded from features set. Finally, significant features were classified by using support vector machine (SVM), K-nearest neighbors (KNN), random forest (RF), and XGBoost classifiers. Due to the class imbalance problem, repeated stratified 5-fold cross-validation was performed in order to tune systems parameters. Results show that among the four above-mentioned models, XGBoost has the best performance for the 5-class classification problem with accuracy: 81.34%{+/-}0.76% and Kappa 0.7388{+/-}0.0101. The proposed model shows promising results, therefore the model can be implemented in Dreem headband to differentiate between sleep states efficiently and be applicable in clinical trial.

Emotion is an intricate physiological response that plays a crucial role in how we respond and cooperate with others in our daily affairs. Numerous experiments have been evolved to recognize emotion, however still require exploration to intensify the performance. To enhance the performance of effective emotion recognition, this study proposes a subject-dependent robust end-to-end emotion recognition system based on a 1D convolutional neural network (1D-CNN). We evaluate the SJTU1 Emotion EEG Dataset SEED-V with five emotions (happy, sad, neural, fear, and disgust). To begin with, we utilize the Fast Fourier Transform (FFT) to decompose the raw EEG signals into six frequency bands and extract the power spectrum feature from the frequency bands. After that, we combine the extracted power spectrum feature with eye movement and differential entropy (DE) features. Finally, for classification, we apply the combined data to our proposed system. Consequently, it attains 99.80% accuracy which surpasses each prior state-of-the-art system.

Electrocardiogram (ECG) is an effective and non-invasive indicator for the detection and prevention of arrhythmia. ECG signals are susceptible to noise contamination, which can lead to errors in ECG interpretation. Therefore, ECG pretreatment is important for accurate analysis. In this paper, a method of noise reduction based on deep learning is proposed. The method is divided into two stages, and two corresponding models are formed. In the first stage, a one-dimensional U-net model is designed for ECG signal denoising to eliminate noise as much as possible. The one-dimensional DR-net model in the second stage is used to reconstruct the ECG signal and to correct the waveform distortion caused by noise removal in the first stage. In this paper, the U-net and the DR-net are constructed by the convolution method to achieve end-to-end mapping from noisy ECG signals to clean ECG signals. The ECG data used in this paper are from CPSC2018, and the noise signal is from MIT-BIH Noise Stress Test Database (NSTDB). In the experiment, the improvement in the signal-to-noise ratio SNRimp, the root mean square error decrease RMSEde, and the correlation coefficient P, are used to evaluate the performance of the network. This two-stage method is compared with FCN and U-net alone. The experimental results show that the two-stage noise reduction method can eliminate complex noise in the ECG signal while retaining the characteristic shape of the ECG signal. According to the results, we believe that the proposed method has a good application prospect in clinical practice.

The main challenge in the clinical assessment of Psychogenic Non-Epileptic Seizures (PNES) is the lack of an electroencephalographic marker in the electroencephalography (EEG) readout. Although decades of EEG studies have focused on detecting cortical brain function underlying PNES, the principle of PNES remains poorly understood. To address this problem, electric potentials generated by large populations of neurons were collected during the resting state to be processed after that by Power Spectrum Density (PSD) for possible analysis of PNES signatures. Additionally, the integration of distributed information of regular and synchronized multi-scale communication within and across inter-regional brain areas has been observed using functional connectivity tools like Phase Lag Index (PLI) and graph-derived metrics. A cohort study of 20 PNES and 19 Healthy Control subjects (HC) were enrolled. The major finding is that PNES patients exhibited significant differences in alpha-power spectrum in brain regions related to cognitive operations, attention, working memory, and movement regulation. Noticeably, we observed that there exists an altered oscillatory activity and a widespread inter-regional phase desynchronization. This indicates changes in global efficiency, node betweenness, shortest path length, and small worldness in the delta, theta, alpha, and beta frequency bands. Finally, our findings look into new evidence of the intrinsic organization of functional brain networks that reflects a dysfunctional level of integration of local activity across brain regions, which can provide new insights into the pathophysiological mechanisms of PNES.

The majority of voice disorders stem from improper vocal usage. Alterations in voice quality can also serve as indicators for a broad spectrum of diseases. Particularly, the significant correlation between voice disorders and dental health underscores the need for precise diagnosis through acoustic data. This paper introduces effective and efficient features for deep learning with speech signals to distinguish between two groups: individuals with healthy voices and those with pathological voice conditions. Using a public voice database, the ten-fold test results obtained from long short-term memory networks trained on the combination of time-frequency and time-space features with a data balance strategy achieved the following metrics: accuracy = 90%, sensitivity = 93%, specificity = 87%, precision = 88%, F1 score = 0.90, and area under the receiver operating characteristic curve = 0.96.

There has been an emerging interest in the study of functional brain networks in cognitive neuroscience in order to better understand brain responses to different stimuli. Such studies can help in understanding brain connectivity alterations that arise in neurodevelopmental disorders such as intellectual disability (ID). This research contributes to this body of knowledge by studying alterations in brain connectivity in ID compared to the typically developing controls (TDC). Electroencephalography (EEG) data of subjects with ID and TDC is collected through limited channel dry electrode system. Data was analyzed for the auditory and rest state processing along with the study of intra-network connectivity of the brain via clustering coefficients. Research findings indicate evidences for links between the sensory deficits and social impairment in ID individuals.

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.

ObjectiveElectroencephalogram (EEG) based motor imagery (MI) classification is an important aspect in brain-machine interfaces (BMIs) which bridges between neural system and computer devices decoding brain signals into recognizable machine commands. However, the MI classification task is challenging due to inherent complex properties, inter-subject variability, and low signal-to-noise ratio (SNR) of EEG signals. To overcome the above-mentioned issues, the current work proposes an efficient multi-scale convolutional neural network (MS-CNN). ApproachIn the framework, discriminant user-specific features have been extracted and integrated to improve the accuracy and performance of the CNN classifier. Additionally, different data augmentation methods have been implemented to further improve the accuracy and robustness of the model. Main resultsThe model achieves an average classification accuracy of 93.74% and Cohens kappa-coefficient of 0.92 on the BCI competition IV2b dataset outperforming several baseline and current state-of-the-art EEG-based MI classification models. SignificanceThe proposed algorithm effectively addresses the shortcoming of existing CNN-based EEG-MI classification models and significantly improves the classification accuracy.

We introduce a simple and interpretable model for classification of electroencephalography (EEG) signals. Our focus essentially is on using deep learning to study how connectivity patterns that are integrated to classify the EEG signals and highlight the important discriminative features used by the model in predictions. In this study, we utilize the connectivity features across different frequency bands in multi edge Graph Neural Networks (GNN) and showed that edge features are complimentary. We use a simple GNN model to predict Frontotemporal Dementia (FTD) in EEG. Our model is capable of achieving average accuracy of approximately 76% using Leave-One-Subject-Out-subject for FTD predictions which are better than the baselines and comparable to State of the arts models. In this article, we study the importance of the connectivity edges, nodes and frequency bands in the prediction of the model, focusing in explainable AI methods through saliency maps to interpret the model both locally and globally. The Saliency maps highlight the importance of Occipital and anterior temporal regions in the prediction of FTD. Furthermore, our results highlight the importance of Alpha and Theta bands in the prediction of FTD. Our observations align with previous research done using classical statistical methods. We argue that there are complimentary information in each each connectivity feature and frequency band brain networks. The impacts of each connectivity metrics on the prediction of the model are quantified to highlight the complimentary information in each connectivity measure.