In recent years, the demand for smart healthcare solutions have heightened the need for accuracy, reliability, and comfort in bedside ECG recording and analysis. This study presents a bedside non-direct contact ECG recording system based on capacitive coupling electrocardiography (cECG) and verifies its performance in accurately capturing Heart Rate Variability (HRV) during the night. Firstly, cECG collects ECG data through clothing, avoiding skin irritation from conventional wet electrodes. Secondly, leveraging the unique characteristics of cECG signals, a deep learning framework assesses the quality of cECG, filtering noise and identifying off-bed information, enhancing HRV analysis precision. Subsequently, the system was employed to recording sleep data from 6 subjects overnight, with our proposed algorithm utilized for signal quality assessment (SQA) and HRV analysis. Finally, HRV features were compared with synchronously collected wet electrode ECG signals, encompassing time domain features, frequency domain features, and nonlinear features, totaling 13 HRV features. Experimental findings demonstrate that for the SQA task, the model achieved a classification accuracy of 94.7%, with a Recall of 0.941, Precision of 0.940, F1 score of 0.941, and Cohen's Kappa of 0.927. The accuracy of on/off-bed monitoring reached 99.79%. Additionally, HRV features showed a strong correlation with the reference ECG. In the time-domain metrics, the largest mean absolute percentage error (MAPE) is for PNN50, with a value of 8.148%. In the frequency-domain features, the largest MAPE is for HF, with a value of 13.253%. For nonlinear features, the largest MAPE is for SD1, with a value of 5.182%. Generally, the system exhibited a reliable solution for cECG recording, on/off-bed status detection, and bedside HRV analysis.

Multimodal learning holds the promise for richer information extraction by capturing dependencies across data sources. Yet, current training methods often underperform due to modality competition, a phenomenon where modalities contend for training resources, leaving some underoptimized. This raises a pivotal question: how can we address training imbalances, ensure adequate optimization across all modalities, and achieve consistent performance improvements as we transition from unimodal to multimodal data? This paper proposes the Multimodal Competition Regularizer (MCR), inspired by a mutual information (MI) decomposition designed to prevent the adverse effects of competition in multimodal training. Our key contributions are: 1) A game-theoretic framework that adaptively balances modality contributions by encouraging each to maximize its informative role in the final prediction. 2) Refining lower and upper bounds for each MI term to enhance the extraction of both task-relevant unique and shared information across modalities. 3) Proposing latent space permutations for conditional MI estimation, significantly improving computational efficiency. MCR outperforms all previously suggested training strategies and simple baselines, demonstrating that training modalities jointly lead to important performance gains on synthetic and large real-world datasets. We release our code and models at https://github.com/kkontras/MCR.

Sleep disorders and their diagnosis are a significant public health concern. Automated sleep stage classification using deep learning models has shown promising results, but these models often lack transparency and interpretability. In this study, we propose an eXplainable Artificial Intelligence (XAI) approach to enhance the interpretability of cutting-edge deep learning sleep stage classification models. The proposed approach consists of a three-steps framework: (i) employing contrastive learning to order a neural network latent space based on input similarity; (ii) mining meaningful instances from that space; and (iii) explaining those instances by a customized XAI methodology. By doing this we are capable of extracting human-comprehensible insights about the model decision-making process, enhancing the applicability of the proposed approach in real-world clinical scenarios. The explanations provided point out high and low-representative sleep epochs of each sleep phase. These sleep epochs are analyzed considering both the single sleep epoch and the sequence of adjacent sleep epochs for the sleep phase classification.Our approach proved to maintain the original model performances, improve the model interpretability, and confirm that the network decision-making process is valid even from the perspective of a physician.

Abstract AI-based emotion recognition approaches may benefit from the integration of multimodal data, but their explainability and validation is still a critical challenge. Indeed, the limited neurophysiological understanding of novel multimodal features, e.g. brain-heart interaction, can be insufficient to assess whether the AI-extracted physiological insights (i.e., the model explanations) accurately reflect the real underlying physiological processes. To validate the explanations obtained by an AI-based model in this context, we introduce a novel framework that autonomously identifies the optimal explanations for a black-box model used in emotion recognition. Our approach leverages a convolutional neural network to process BHI features, which are derived from EEG and HRV data and rearranged as images. A model-agnostic methodology is employed to extract local explanations, which are then dynamically evaluated to select the most accurate for representing specific emotional states. The effectiveness of the proposed framework is evaluated across multiple classification tasks, including up to 9-level arousal and valence emotion classification, as well as nine discrete emotions classification, using the MAHNOB-HCI and DEAP datasets. The system achieved remarkable accuracy levels, consistently reaching approximately 97–98% across all tasks. Furthermore, our dynamic selection framework revealed that Integrated Gradients outperformed other state-of-the-art explainable AI approaches in reliably capturing global explanations.

Chronic widespread pain remains a complex and incompletely understood condition. To complement existing pain assessment strategies, this study explored the ecological validity of unobtrusively captured daily life physiological signals as indicators of pain. Therefore, we collected physiological data using a wearable wristband from 46 patients with chronic widespread pain for seven days. Linear mixed-effect models revealed several significant associations between physiological signals, such as mean heart rate and momentary pain intensity. However, making individual pain predictions with multivariate machine learning models did not add value. While this study underscores the potential of ambulatory physiology for pain assessment, future research should validate and expand upon these initial findings to further enhance pain management strategies.

OBJECTIVE: Building large-scale data bases of biomedical signal recordings for training artificial-intelligence systems involves substantial human effort in data processing and annotation. In the case of event detection, experts need to exhaustively scroll through the recordings and highlight events of interest. METHODS: We propose an iterative annotation support algorithm with a human in the loop to improve the efficiency of the annotation process. Our algorithm generates proposal events based on an event detection model trained on incomplete annotations. The human only needs to verify candidate events proposed by the tool instead of scrolling through the entire data set. Our algorithm iterates between proposal generation and verification to leverage the human-in-the-loop feedback to obtain a growing set of event annotations. RESULTS: Our algorithm finds a substantial amount of events at a fraction of the human time spent when comparing with a benchmark method and the normal manual process, finding all events in one data set and 70% of events in another with the human-in-the-loop only viewing 20% of the data. CONCLUSION: Our results show that combining human and computer effort can substantially speed up the annotation process for events in biomedical signal processing. SIGNIFICANCE: Due to its simplicity and minimal reliance on task-specific information, our algorithm is broadly applicable, unlocking substantial improvements in the scalability and efficiency of biomedical signal annotation.

Recent advances in deep learning (DL) have significantly impacted motor imagery (MI)-based brain-computer interface (BCI) systems, enhancing the decoding of electroencephalography (EEG) signals. However, most studies struggle to identify discriminative patterns across subjects during MI tasks, limiting MI classification performance. In this paper, we propose MixNet, a novel classification framework designed to overcome this limitation by utilizing spectral-spatial signals from MI data, along with a multi-task learning architecture named MIN2Net, for classification. Here, the spectral-spatial signals are generated using the filter-bank common spatial patterns (FBCSP) method on MI data. Since the multi-task learning architecture is used for the classification task, the learning in each task may exhibit different generalization rates and potential overfitting across tasks. To address this issue, we implement adaptive gradient blending, simultaneously regulating multiple loss weights and adjusting the learning pace for each task based on its generalization/overfitting tendencies. Experimental results on six benchmark datasets of different data sizes demonstrate that MixNet consistently outperforms all state-of-the-art algorithms in subject-dependent and -independent settings. Finally, the low-density EEG-MI classification results show MixNet’s superiority over state-of-the-art algorithms, offering promising implications for Internet of Thing (IoT) applications such as lightweight and portable EEG wearable devices based on low-density montages.

Wearable EEG enables us to capture large amounts of high-quality sleep data for diagnostic purposes. To make full use of this capacity we need high-performance automatic sleep scoring models. To this end, it has been noted that domain mismatch between recording equipment can be considerable, e.g. PSG to wearable EEG, but a previously observed benefit from personalizing models to individual subjects further indicates a personal domain in sleep EEG. In this work, we have investigated the extent of such a personal domain in wearable EEG, and review supervised and unsupervised approaches to personalization as found in the literature. We investigated the personalization effect of the unsupervised Adversarial Domain Adaptation and implemented an unsupervised method based on statistics alignment. No beneficial personalization effect was observed using these unsupervised methods. We find that supervised personalization leads to a substantial performance improvement on the target subject ranging from 15% Cohen's Kappa for subjects with poor performance ( ) to roughly 2% on subjects with high performance ( ). This improvement was present for models trained on both small and large data sets, indicating that even high-performance models benefit from supervised personalization. We found that this personalization can be beneficially regularized using Kullback-Leibler regularization, leading to lower variance with negligible cost to improvement. Based on the experiments, we recommend model personalization using Kullback-Leibler regularization.

The idea of a systematic digital representation of the entire known human pathophysiology, which we could call the Virtual Human Twin, has been around for decades. To date, most research groups focused instead on developing highly specialised, highly focused patient-specific models able to predict specific quantities of clinical relevance. While it has facilitated harvesting the low-hanging fruits, this narrow focus is, in the long run, leaving some significant challenges that slow the adoption of digital twins in healthcare. This position paper lays the conceptual foundations for developing the Virtual Human Twin (VHT). The VHT is intended as a distributed and collaborative infrastructure, a collection of technologies and resources (data, models) that enable it, and a collection of Standard Operating Procedures (SOP) that regulate its use. The VHT infrastructure aims to facilitate academic researchers, public organisations, and the biomedical industry in developing and validating new digital twins in healthcare solutions with the possibility of integrating multiple resources if required by the specific context of use. Healthcare professionals and patients can also use the VHT infrastructure for clinical decision support or personalised health forecasting. As the European Commission launched the EDITH coordination and support action to develop a roadmap for the development of the Virtual Human Twin, this position paper is intended as a starting point for the consensus process and a call to arms for all stakeholders.

Tonic-clonic seizures (TCSs) pose a significant risk for sudden unexpected death in epilepsy (SUDEP). Previous research has highlighted the potential of multimodal wearable seizure detection systems in accurately detecting TCSs through continuous monitoring, enabling timely alarms and potentially preventing SUDEP. However, such multimodal systems carry a higher risk of sensor malfunction. In this paper, we propose a cyclic transformer approach to address these challenges. The cyclic transformer learns a robust representation by performing circular modal translations between the source and target modalities. It leverages back-translation as regularization technique to enhance the discriminative power of the learned representation. Notably, the proposed cyclic transformer is trained on paired multimodal data but requires only a single source modality during deployment. This characteristic ensures the robustness of the cyclic transformer to perturbations or missing information in the target modality. Experimental results demonstrate that the proposed cyclic transformer achieves competitive performance compared with existing multimodal systems. While both approaches were trained using EEG and EMG data, the cyclic transformer exclusively employs EEG data for testing, diverging from the state-of-the-art's utilization of both EEG and EMG data during test. This showcases the effectiveness of the cyclic transformer in multimodal TCSs detection, offering a promising approach for enhancing the accuracy and robustness of seizure detection systems while mitigating the risks associated with sensor malfunction.

Change Point Detection (CPD) refers to the task of identifying abrupt changes in the characteristics or statistics of time series data. Recent advancements have led to a shift away from traditional model-based CPD approaches, which rely on predefined statistical distributions, toward neural network-based and distribution-free methods using autoencoders. However, many state-of-the-art methods in this category often neglect to explicitly leverage spatial information across multiple channels, making them less effective at detecting changes in cross-channel statistics. In this paper, we introduce an unsupervised, distribution-free CPD method that explicitly incorporates both temporal and spatial (cross-channel) information in multi-channel time series data based on the so-called Time-Invariant Representation (TIRE) autoencoder. Our evaluation, conducted on both simulated and real-life datasets, illustrates the significant advantages of our proposed multi-channel TIRE (MC-TIRE) method, which consistently delivers more accurate CPD results.

In this paper, we describe the epilepsy detection grand challenge, in association with ICASSP 2023. The challenge was centered on seizure detection using wearable behind-the-ear EEG. Two separate tasks were set for the participants: 1) obtain the best overall performance in seizure detection; 2) systematically engineer the data in a data-centric task to optimize a given AI model for seizure detection.

We propose a regularization method, so-called temporal smoothness regularization, for training deep neural networks for automatic sleep staging in small data settings. In intuition, we constrain the cross-entropy losses of any two adjacent epochs in the sequential input to be as close to each other as possible. The regularization closely reflects the slow transition nature of sleep process which implies small information changes between two consecutive sleep epochs. Via the regularization, we essentially discourage the network from overfitting to these small changes. Our experiments show that training the SeqSleepNet base network with the proposed regularization leads to performance improvement over the baseline without the regularization applied. Furthermore, our developed method achieves the performance on par with the state-of-the-art performance while outperforming other existing methods.

Change point detection (CPD) refers to the problem of detecting changes in the statistics of pseudo-stationary signals or time series. A recent trend in CPD research is to replace the traditional statistical tests with distribution-free autoencoder-based algorithms, which can automatically learn complex patterns in time series data. In particular, the so-called time-invariant representation (TIRE) models have gained traction, as these separately encode time-variant and time-invariant subfeatures, as opposed to traditional autoencoders. However, optimizing the trade-off between two loss terms, i.e., the reconstruction loss and the time-invariant loss, is challenging. To address this issue, we propose a novel loss function that elegantly combines both losses without the need for manually tuning a trade-off hyperparameter. We demonstrate that this new hyperparameter-free loss, in combination with a relatively simple convolutional neural network (CNN), consistently achieves superior or comparable performance compared to the manually-tuned baseline TIRE models across diverse benchmark datasets, both simulated and real-life. In addition, we present a representation analysis, demonstrating that the distribution of the time-invariant features extracted by our model is more concentrated within the same segment (more so than with previous TIRE models), which implies that these features can potentially be used for other applications, such as classification and clustering.

Personalized longitudinal disease assessment is central to quickly diagnosing, appropriately managing, and optimally adapting the therapeutic strategy of multiple sclerosis (MS). It is also important for identifying idiosyncratic subject-specific disease profiles. Here, we design a novel longitudinal model to map individual disease trajectories in an automated way using smartphone sensor data that may contain missing values. First, we collect digital measurements related to gait and balance, and upper extremity functions using sensor-based assessments administered on a smartphone. Next, we treat missing data via imputation. We then discover potential markers of MS by employing a generalized estimation equation. Subsequently, parameters learned from multiple training datasets are ensembled to form a simple, unified longitudinal predictive model to forecast MS over time in previously unseen people with MS. To mitigate potential underestimation for individuals with severe disease scores, the final model incorporates additional subject-specific fine-tuning using data from the first day. The results show that the proposed model is promising to achieve personalized longitudinal MS assessment; they also suggest that features related to gait and balance as well as upper extremity function, remotely collected from sensor-based assessments, may be useful digital markers for predicting MS over time.

Human sleep is cyclical with a period of approximately 90 minutes, implying long temporal dependency in the sleep data. Yet, exploring this long-term dependency when developing sleep staging models has remained untouched. In this work, we show that while encoding the logic of a whole sleep cycle is crucial to improve sleep staging performance, the sequential modelling approach in existing state-of-the-art deep learning models are inefficient for that purpose. We thus introduce a method for efficient long sequence modelling and propose a new deep learning model, L-SeqSleepNet, which takes into account whole-cycle sleep information for sleep staging. Evaluating L-SeqSleepNet on four distinct databases of various sizes, we demonstrate state-of-the-art performance obtained by the model over three different EEG setups, including scalp EEG in conventional Polysomnography (PSG), in-ear EEG, and around-the-ear EEG (cEEGrid), even with a single EEG channel input. Our analyses also show that L-SeqSleepNet is able to alleviate the predominance of N2 sleep (the major class in terms of classification) to bring down errors in other sleep stages. Moreover the network becomes much more robust, meaning that for all subjects where the baseline method had exceptionally poor performance, their performance are improved significantly. Finally, the computation time only grows at a sub-linear rate when the sequence length increases.

Automating sleep staging is vital to scale up sleep assessment and diagnosis to serve millions experiencing sleep deprivation and disorders and enable longitudinal sleep monitoring in home environments. Learning from raw polysomnography signals and their derived time-frequency image representations has been prevalent. However, learning from multi-view inputs (e.g., both the raw signals and the time-frequency images) for sleep staging is difficult and not well understood. This work proposes a sequence-to-sequence sleep staging model, XSleepNet,1that is capable of learning a joint representation from both raw signals and time-frequency images. Since different views may generalize or overfit at different rates, the proposed network is trained such that the learning pace on each view is adapted based on their generalization/overfitting behavior. In simple terms, the learning on a particular view is speeded up when it is generalizing well and slowed down when it is overfitting. View-specific generalization/overfitting measures are computed on-the-fly during the training course and used to derive weights to blend the gradients from different views. As a result, the network is able to retain the representation power of different views in the joint features which represent the underlying distribution better than those learned by each individual view alone. Furthermore, the XSleepNet architecture is principally designed to gain robustness to the amount of training data and to increase the complementarity between the input views. Experimental results on five databases of different sizes show that XSleepNet consistently outperforms the single-view baselines and the multi-view baseline with a simple fusion strategy. Finally, XSleepNet also outperforms prior sleep staging methods and improves previous state-of-the-art results on the experimental databases.

In this paper, we introduce a new variation of the Convolutional Neural Network Inception block, called Sinc, for sleep stage classification in premature newborn babies using electroencephalogram (EEG). In practice, there are many medical centres where only a limited number of EEG channels are recorded. Existing automated algorithms mainly use multi-channel EEGs which perform poorly when fewer numbers of channels are available. The proposed Sinc utilizes multi-scale analysis to place emphasis on the temporal EEG information to be less dependent on the number of EEG channels. In Sinc, we increase the receptive fields through Inception while by additionally sharing the filters that have similar receptive fields, overfitting is controlled and the number of trainable parameters dramatically reduced. To train and test this model, 96 longitudinal EEG recordings from 26 premature infants are used. The Sinc-based model significantly outperforms state-of-the-art neonatal quiet sleep detection algorithms, with mean Kappa 0.77 ± 0.01 (with 8-channel EEG) and 0.75 ± 0.01 (with a single bipolar channel EEG). This is the first study using Inception-based networks for EEG analysis that utilizes filter sharing to improve efficiency and trainability. The suggested network can successfully detect quiet sleep stages with even a single EEG channel making it more practical especially in the hospital setting where cerebral function monitoring is predominantly used.

Although the development of computer-aided algorithms for sleep staging is integrated into automatic detection of sleep disorders, most supervised deep learning-based models might suffer from insufficient labeled data. While the adoption of semi-supervised learning (SSL) can mitigate the issue, the SSL models are still limited to the lack of discriminative feature extraction for diverse obstructive sleep apnea (OSA) severity. This model deterioration might be exacerbated during the domain adaptation. Such exploration on the alleviation of domain-shift of SSL model between different OSA conditions has attracted more and more attentions from the clinic. In this work, a co-attention meta sleep staging network (CMS2-net) is proposed to simultaneously deal with two issues: the inter-class disparity problem and the intra-class selection problem. Within CMS2-net, a co-attention module and a triple-classifier are designed to explicitly refine the coarse feature representations by identifying the class boundary inconsistency. Moreover, the mutual information with meta contrastive variance is introduced to supervise the gradient stream from a multi-scale view. The performance of the proposed framework is demonstrated on both public and local datasets. Furthermore, our approach achieves the state-of-the-art SSL results on both datasets.

Leveraging consumer technology such as smartphone and smartwatch devices to objectively assess people with multiple sclerosis (PwMS) remotely could capture unique aspects of disease progression. This study explores the feasibility of assessing PwMS and Healthy Control's (HC) physical function by characterising gaitrelated features, which can be modelled using machine learning (ML) techniques to correctly distinguish subgroups of PwMS from healthy controls. A total of 97 subjects (24 HC subjects, 52 mildly disabled (PwMSmild, EDSS [0-3]) and 21 moderately disabled (PwMSmod, EDSS [3.5- 5.5]) contributed data which was recorded from a TwoMinute Walk Test (2MWT) performed out-of-clinic and daily over a 24-week period. Signal-based features relating to movement were extracted from sensors in smartphone and smartwatch devices. A large number of features (n = 156) showed fair-to-strong (R > 0.3) correlations with clinical outcomes. LASSO feature selection was applied to select and rank subsets of features used for dichotomous classification between subject groups, which were compared using Logistic Regression (LR), Support Vector Machines (SVM) and Random Forest (RF) models. Classifications of subject types were compared using data obtained from smartphone, smartwatch and the fusion of features from both devices. Models built on smartphone features alone achieved the highest classification performance, indicating that accurate and remote measurement of the ambulatory characteristics of HC and PwMS can be achieved with only one device. It was observed however that smartphonebased performance was affected by inconsistent placement location (running belt versus pocket). Results show that PwMSmod could be distinguished from HC subjects (Acc. 82.2 ± 2.9%, Sen. 80.1 ± 3.9%, Spec. 87.2 ± 4.2%, F184.3 ± 3.8), and PwMSmild (Acc. 82.3 ± 1.9%, Sen. 71.6 ± 4.2%, Spec. 87.0 ± 3.2%, F1 75.1 ± 2.2) using an SVM classifier with a Radial Basis Function (RBF). PwMSmild were shown to exhibit HC-like behaviour and were thus less distinguishable from HC (Acc. 66.4 ± 4.5%, Sen. 67.5 ± 5.7%, Spec. 60.3 ± 6.7%, F158.6 ± 5.8). Finally, it was observed that subjects in this study demonstrated low intraand high inter-subject variability which was representative of subject-specific gait characteristics.

Generative adversarial networks (GAN) have recently been shown to be efficient for speech enhancement. However, most, if not all, existing speech enhancement GANs (SEGAN) make use of a single generator to perform one-stage enhancement mapping. In this work, we propose to use multiple generators that are chained to perform multi-stage enhancement mapping, which gradually refines the noisy input signals in a stage-wise fashion. Furthermore, we study two scenarios: (1) the generators share their parameters and (2) the generators' parameters are independent. The former constrains the generators to learn a common mapping that is iteratively applied at all enhancement stages and results in a small model footprint. On the contrary, the latter allows the generators to flexibly learn different enhancement mappings at different stages of the network at the cost of an increased model size. We demonstrate that the proposed multi-stage enhancement approach outperforms the one-stage SEGAN baseline, where the independent generators lead to more favorable results than the tied generators. The source code is available at http://github.com/pquochuy/idsegan.

We propose a multi-label multi-task framework based on a convolutional recurrent neural network to unify detection of isolated and overlapping audio events. The framework leverages the power of convolutional recurrent neural network architectures; convolutional layers learn effective features over which higher recurrent layers perform sequential modelling. Furthermore, the output layer is designed to handle arbitrary degrees of event overlap. At each time step in the recurrent output sequence, an output triple is dedicated to each event category of interest to jointly model event occurrence and temporal boundaries. That is, the network jointly determines whether an event of this category occurs, and when it occurs, by estimating onset and offset positions at each recurrent time step. We then introduce three sequential losses for network training: multi-label classification loss, distance estimation loss, and confidence loss. We demonstrate good generalization on two datasets: ITC-Irst for isolated audio event detection, and TUT-SED-Synthetic-2016 for overlapping audio event detection.

Multi-source datasets often present the challenge of source-wise missing data which can render large portions of the dataset inaccessible. The applicability of traditional missing data techniques on multi-source datasets is poorly understood. We present the first quantitative evaluation of the state-of-the-art missing data techniques as applied to a freely available dataset of smart-phone recordings from Parkinsonian patients wherein source-wise missing data is simulated. The classification accuracy and imputation error of five missing data techniques, including a multi-modal autoencoder and multi-source ensemble learning, are compared at varying levels of missingness. These results demonstrate the relative applicability of each technique under different conditions and subsequently highlight the challenges of source-wise missing on remotely collected datasets. Specifically, multi-source ensemble learning proves to be a highly successful alternative to the traditional imputation techniques when a majority of observations possess missing data.

Acoustic scenes are rich and redundant in their content. In this work, we present a spatio-temporal attention pooling layer coupled with a convolutional recurrent neural network to learn from patterns that are discriminative while suppressing those that are irrelevant for acoustic scene classification. The convolutional layers in this network learn invariant features from time-frequency input. The bidirectional recurrent layers are then able to encode the temporal dynamics of the resulting convolutional features. Afterwards, a two-dimensional attention mask is formed via the outer product of the spatial and temporal attention vectors learned from two designated attention layers to weigh and pool the recurrent output into a final feature vector for classification. The network is trained with between-class examples generated from between-class data augmentation. Experiments demonstrate that the proposed method not only outperforms a strong convolutional neural network baseline but also sets new state-of-the-art performance on the LITIS Rouen dataset.

Identifying a core set of features is one of the most important steps in the development of an automated seizure detector. In most of the published studies describing features and seizure classifiers, the features were hand-engineered, which may not be optimal. The main goal of the present paper is using deep convolutional neural networks (CNNs) and random forest to automatically optimize feature selection and classification. The input of the proposed classifier is raw multi-channel EEG and the output is the class label: seizure/nonseizure. By training this network, the required features are optimized, while fitting a nonlinear classifier on the features. After training the network with EEG recordings of 26 neonates, five end layers performing the classification were replaced with a random forest classifier in order to improve the performance. This resulted in a false alarm rate of 0.9 per hour and seizure detection rate of 77% using a test set of EEG recordings of 22 neonates that also included dubious seizures. The newly proposed CNN classifier outperformed three data-driven feature-based approaches and performed similar to a previously developed heuristic method.

In neonatal intensive care units, there is a need for around the clock monitoring of electroencephalogram (EEG), especially for recognizing seizures. An automated seizure detector with an acceptable performance can partly fill this need. In order to develop a detector, an extensive dataset labeled by experts is needed. However, accurately defining neonatal seizures on EEG is a challenge, especially when seizure discharges do not meet exact definitions of repetitiveness or evolution in amplitude and frequency. When several readers score seizures independently, disagreement can be high. Commonly used metrics such as good detection rate (GDR) and false alarm rate (FAR) derived from data scored by multiple raters have their limitations. Therefore, new metrics are needed to measure the performance with respect to the different labels. In this paper, instead of defining the labels by consensus or majority voting, popular metrics including GDR, FAR, positive predictive value, sensitivity, specificity, and selectivity are modified such that they can take different scores into account. To this end, 353 hours of EEG data containing seizures from 81 neonates were visually scored by a clinical neurophysiologist, and then processed by an automated seizure detector. The scored seizures were mixed with false detections of an automated seizure detector and were relabeled by three independent EEG readers. Then, all labels were used in the proposed performance metrics and the result was compared with the majority voting technique and showed higher accuracy and robustness for the proposed metrics. Results were confirmed using a bootstrapping test.

Sleep state development in preterm neonates can provide crucial information regarding functional brain maturation and give insight into neurological well being. However, visual labeling of sleep stages from EEG requires expertise and is very time consuming, prompting the need for an automated procedure. We present a robust method for automated detection of preterm sleep from EEG, over a wide postmenstrual age ([Formula: see text] age) range, focusing first on Quiet Sleep (QS) as an initial marker for sleep assessment. Our algorithm, CLuster-based Adaptive Sleep Staging (CLASS), detects QS if it remains relatively more discontinuous than non-QS over PMA. CLASS was optimized on a training set of 34 recordings aged 27-42 weeks PMA, and performance then assessed on a distinct test set of 55 recordings of the same age range. Results were compared to visual QS labeling from two independent raters (with inter-rater agreement [Formula: see text]), using Sensitivity, Specificity, Detection Factor ([Formula: see text] of visual QS periods correctly detected by CLASS) and Misclassification Factor ([Formula: see text] of CLASS-detected QS periods that are misclassified). CLASS performance proved optimal across recordings at 31-38 weeks (median [Formula: see text], median MF 0-0.25, median Sensitivity 0.93-1.0, and median Specificity 0.80-0.91 across this age range), with minimal misclassifications at 35-36 weeks (median [Formula: see text]). To illustrate the potential of CLASS in facilitating clinical research, normal maturational trends over PMA were derived from CLASS-estimated QS periods, visual QS estimates, and nonstate specific periods (containing QS and non-QS) in the EEG recording. CLASS QS trends agreed with those from visual QS, with both showing stronger correlations than nonstate specific trends. This highlights the benefit of automated QS detection for exploring brain maturation.

To decode auditory attention from electroencephalography (EEG) recordings in a cocktail-party scenario with two competing speakers a least-squares method has recently been proposed, showing a promising decoding accuracy. This method however requires the clean speech signals of both the attended and the unattended speaker to be available as reference signals, which is difficult to achieve from the noisy recorded microphone signals in practice. In addition, optimizing the parameters involved in the spatio-temporal filter design is of crucial importance in order to reach the largest possible decoding performance. In this paper, the influence of noisy acoustic reference signals and the spatio-temporal filter and regularization parameters on the decoding performance is investigated. The results show that to some extent the decoding performance is robust to noisy acoustic reference signals, depending on the noise type. Furthermore, we demonstrate the crucial influence of several parameters on the decoding performance, especially when the acoustic reference signals used for decoding have been corrupted by noise.

The automated analysis of the EEG pattern of the preterm newborn would be a valuable tool in the neonatal intensive care units for the prognosis of neurological development. The analysis of the (a)symmetry between the two hemispheres can provide useful information about neuronal dysfunction in early stages. Consecutive and subgroup analyses of different brain regions will allow to detect physiologic asymmetry versus pathologic asymmetry. This can improve the assessment of the long-term neurodevelopmental outcome. We show that pathological asymmetry can be measured and detected using the channel symmetry index, which comprises the difference in power spectral density of contralateral EEG signals. To distinguish pathological from physiological normal EEG patterns, we make use of one-class SVM classifiers. Future work will focus on adding relevant features for classification and augmenting the sample size, thus reducing the overall classification error.

We propose a new algorithm to impose independence constraints based on second order statistics in one mode of the Parallel Factor Analysis / Canonical Decomposition, also known as the CP model, and show with simulations that it outperforms in some cases the ordinary CP model.