High mental workload poses significant challenges across various domains, often impairing performance and decision-making. Recent advancements in Internet of Things (IoT) technologies have enabled continuous monitoring of brain activity via electroencephalography (EEG), facilitating real-time assessment of mental workload during task execution. Early detection of elevated workload levels is essential for mitigating potential adverse outcomes. This paper presents EEGMeNet, an end-to-end, multi-task neural network designed to classify mental workload levels from EEG signals. EEGMeNet consists of four key components: (1) a local feature learner that extracts spectral and spatial characteristics, (2) a global feature learner employing attention-based mechanisms to model temporal dependencies, (3) a feature preserver that enhances representation learning, and (4) a supervised classifier for mental workload classification. The proposed model is evaluated across five mental task EEG datasets—including one acquired in clinical settings—under both within-subject and cross-subject, as well as within-session and cross-session classification paradigms. EEGMeNet outperforms nine state-of-the-art baselines, achieving an accuracy of 87.49±2.66% and an F1 score of 85.26±2.63% in cross-subject evaluations, showing at least a 9% improvement in accuracy and a 12% gain in F1 score. Moreover, EEGMeNet demonstrates low prediction time complexity, making it well-suited for deployment on resource-constrained IoT devices. These results underscore EEGMeNet’s robustness to inter-subject and multi-session EEG variability, validating its effectiveness for real-world EEG-IoT applications. This work establishes EEGMeNet as a promising foundation for future mental workload classification research in EEG-IoT environments.

In this study, we propose AlphaGrad, a novel adaptive loss blending strategy for optimizing multi-task learning (MTL) models in motor imagery (MI)-based electroencephalography (EEG) classification. AlphaGrad is the first method to automatically adjust multi-loss functions with differing metric scales, including mean square error, cross-entropy, and deep metric learning, within the context of MI-EEG. We evaluate AlphaGrad using two state-of-the-art MTL-based neural networks, MIN2Net and FBMSNet, across four benchmark datasets. Experimental results show that AlphaGrad consistently outperforms existing strategies such as AdaMT, GradApprox, and fixed-weight baselines in classification accuracy and training stability. Compared to baseline static weighting, AlphaGrad achieves over 10% accuracy improvement on subject-independent MI tasks when evaluated on the largest benchmark dataset. Furthermore, AlphaGrad demonstrates robust adaptability across various EEG paradigms-including steady-state visually evoked potential (SSVEP) and event-related potential (ERP), making it broadly applicable to brain-computer interface (BCI) systems. We also provide gradient trajectory visualizations highlighting AlphaGrad's ability to maintain training stability and avoid local minima. These findings underscore AlphaGrad's promise as a general-purpose solution for adaptive multi-loss optimization in biomedical time-series learning.

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.

Detecting obstructive sleep apnea (OSA) is essential for diagnosing and managing sleep health. Traditionally, this involves clinical settings with hardly accessible processes. We propose that the automated detection of OSA events is achievable using features extracted from fingertip photoplethysmography (PPG) signals combined with modern deep learning (DL) techniques. Utilizing two benchmark data sets with extensive PPG recordings, we introduce ApSense, a DL model designed for the OSA event onset recognition from PPG features. ApSense presents a custom neural architecture and domain-specific feature extraction from PPG waveforms. We benchmark it against the state-of-the-art (SOTA) algorithms, including RRWaveNet, PPGNetSA, AIOSA, DRIVEN, and LeNet-5. In our evaluations, ApSense demonstrated improved sensitivity, specificity, and area under the receiver operating characteristic (AUROC) on the test data sets. Furthermore, an ablation study highlighted strategic customizations of ApSense, enhancing its performance and adaptability to different data sets. ApSense demonstrates high reliability, as its outstanding results were confirmed even in high-variance data sets. By detecting OSA events, ApSense enables the estimation of the predicted apnea-hypopnea index (pAHI), which can be used for prescreening individuals for sleep apnea in a low-cost setup. ApSense shows the potential for the PPG-based OSA detection and clinical applications for prescreening in the future.

In light of the COVID-19 pandemic, patients were required to manually input their daily oxygen saturation (SpO2) and pulse rate (PR) values into a health monitoring system—unfortunately, such a process trend to be an error in typing. Several studies attempted to detect the physiological value from the captured image using optical character recognition (OCR). However, the technology has limited availability with high cost. Thus, this study aimed to propose a novel framework called pandemic accelerated human-machine collaboration (PACMAN) with a low-resource deep learning-based computer vision. We compared state-of-the-art object detection algorithms (scaled YOLOv4, YOLOv5, and YOLOR), including the commercial OCR tools for digit recognition on the captured images from the pulse oximeter display. All images were derived from crowdsourced data collection with varying quality and alignment. YOLOv5 was the best performing model against the given model comparison across all data sets, notably the correctly orientated image data set. We further improved the model performance with the digits auto-orientation algorithm and applied a clustering algorithm to extract SpO2and PR values. The accuracy performance of YOLOv5 with the implementations was approximately 81.0–89.5%, which was enhanced compared to without any additional implementation. Accordingly, this study highlighted the completion of the PACMAN framework to detect and read digits in real-world data sets. The proposed framework has been currently integrated into the patient monitoring system utilized by hospitals nationwide.

Respiratory rate (RR) is an important biomarker as RR changes can reflect severe medical events, such as heart disease, lung disease, and sleep disorders. Unfortunately, standard manual RR counting is prone to human error and cannot be performed continuously. This study proposes a method for continuously estimating RR, RRWaveNet. The method is a compact end-to-end deep learning model which does not require feature engineering and can use low-cost raw photoplethysmography (PPG) as input signal. RRWaveNet was tested subject-independently and compared to baseline in four data sets (BIDMC, CapnoBase, WESAD, and SensAI) and using three window sizes (16, 32, and 64 s). RRWaveNet outperformed current state-of-the-art methods with mean absolute errors at an optimal window size of 1.66 ± 1.01, 1.59 ± 1.08, 1.92 ± 0.96, and 1.23 ± 0.61 breaths per minute for each data set. In remote monitoring settings, such as in the WESAD and SensAI data sets, we apply transfer learning to improve the performance using two other ICU data sets as pretraining data sets, reducing the MAE by up to 21%. This shows that this model allows accurate and practical estimation of RR on affordable and wearable devices. Our study also shows feasibility of remote RR monitoring in the context of telemedicine and at home.

We propose a deep learning-based approach to localizing a small moving object with a single millimeter-wave frequency-modulated continuous-wave (FMCW) radar. The main challenge that foils conventional localization techniques, such as 3-D fast Fourier transform (3-D-FFT), Pisarenko method, multiple signal classification (MUSIC), estimation of signal parameters via rotational invariance technique (ESPRIT), Capon’s method, and Burg’s method, is the low signal-to-noise ratio of the reflected signal from millimeter-sized objects. Our key idea is to combine useful but noisy features from classical transforms [e.g., fast Fourier transform (FFT)] with neural networks that can refine and interpret those features into range and angle estimates by training on a large dataset of examples. Importantly, our networks were designed to be translation-equivariant, which enables accurate predictions of unseen object locations and improves the range and azimuth root mean square error (RMSE) scores by 34%–46% and 41%–60%, respectively, over state-of-the-art approaches. This pilot study establishes a new baseline for small-object tracking using FMCW and can enable tracking of small animals, such as ants inside the colony for behavior studies. Our first FMCW-small-object dataset and the source code are publicly available onhttps://github.com/shikuzen/RA-CNN.

The detection of seizures in epileptic patients via Electroencephalography (EEG) is an essential key to medical treatment. With the advances in deep learning, many approaches are proposed to tackle this problem. However, concerns such as performance, speed, and subject-independency should still be considered for practical application. Thus, we propose EEGWaveNet, a novel end-to-end multiscale convolutional neural network designed to address epileptic seizure detection. Our network utilizes trainable depth-wise convolutions as discriminative filters to simultaneously gather features from each EEG channel and separate the signal into multiscale resolution. Then, the spatial-temporal features are extracted from each scale for further classification. To demonstrate the effectiveness of EEGWaveNet, we evaluate the model in three datasets: CHB-MIT, TUSZ, and BONN. From the results, EEGWaveNet’s performance is comparable to other baseline methods in the subject-dependent approach and outperforms the others in subject-independent approaches. EEGWaveNet also has time complexity comparable to the compact EEGNet-8,2. Moreover, we transfer the model trained from the subject-independent approach and fine-tune it with a 1-h recording, significantly improving sensitivity and F1-score (Binary) compared to without fine-tuning. This article indicates the possibility of further developing this model and the fine-tuning methodology toward healthcare 5.0, where the AI aid clinicians in a manner of man–machine collaboration.

The elimination of ocular artifacts is critical in analyzing electroencephalography (EEG) data for various brain-computer interface (BCI) applications. Despite numerous promising solutions, electrooculography (EOG) recording or an eye-blink detection algorithm is required for the majority of artifact removal algorithms. This reliance can hinder the model's implementation in real-world applications. This paper proposes EEGANet, a framework based on generative adversarial networks (GANs), to address this issue as a data-driven assistive tool for ocular artifacts removal (source code is available at https://github.com/IoBT-VISTEC/EEGANet). After the model was trained, the removal of ocular artifacts could be applied calibration-free without relying on the EOG channels or the eye blink detection algorithms. First, we tested EEGANet's ability to generate multi-channel EEG signals, artifacts removal performance, and robustness using the EEG eye artifact dataset, which contains a significant degree of data fluctuation. According to the results, EEGANet is comparable to state-of-the-art approaches that utilize EOG channels for artifact removal. Moreover, we demonstrated the effectiveness of EEGANet in BCI applications utilizing two distinct datasets under inter-day and subject-independent schemes. Despite the absence of EOG signals, the classification performance of the signals processed by EEGANet is equivalent to that of traditional baseline methods. This study demonstrates the potential for further use of GANs as a data-driven artifact removal technique for any multivariate time-series bio-signal, which might be a valuable step towards building next-generation healthcare technology.

Identifying bio-signals based-sleep stages requires time-consuming and tedious labor of skilled clinicians. Deep learning approaches have been introduced in order to challenge the automatic sleep stage classification conundrum. However, the difficulties can be posed in replacing the clinicians with the automatic system due to the differences in many aspects found in individual bio-signals, causing the inconsistency in the performance of the model on every incoming individual. Thus, we aim to explore the feasibility of using a novel approach, capable of assisting the clinicians and lessening the workload. We propose the transfer learning framework, entitled MetaSleepLearner, based on Model Agnostic Meta-Learning (MAML), in order to transfer the acquired sleep staging knowledge from a large dataset to new individual subjects (source code is available at https://github.com/IoBT-VISTEC/MetaSleepLearner). The framework was demonstrated to require the labelling of only a few sleep epochs by the clinicians and allow the remainder to be handled by the system. Layer-wise Relevance Propagation (LRP) was also applied to understand the learning course of our approach. In all acquired datasets, in comparison to the conventional approach, MetaSleepLearner achieved a range of 5.4% to 17.7% improvement with statistical difference in the mean of both approaches. The illustration of the model interpretation after the adaptation to each subject also confirmed that the performance was directed towards reasonable learning. MetaSleepLearner outperformed the conventional approaches as a result from the fine-tuning using the recordings of both healthy subjects and patients. This is the first work that investigated a non-conventional pre-training method, MAML, resulting in a possibility for human-machine collaboration in sleep stage classification and easing the burden of the clinicians in labelling the sleep stages through only several epochs rather than an entire recording.

Recognizing movements during sleep is crucial for the monitoring of patients with sleep disorders, and the utilization of ultra-wideband (UWB) radar for the classification of human sleep postures has not been explored widely. This study investigates the performance of an off-the-shelf single antenna UWB in a novel application of sleep postural transition (SPT) recognition. The proposed Multi-View Learning, entitled SleepPoseNet or SPN, with time series data augmentation aims to classify four standard SPTs. SPN exhibits an ability to capture both time and frequency features, including the movement and direction of sleeping positions. The data recorded from 38 volunteers displayed that SPN with a mean accuracy of 73.7 ±0.8 % significantly outperformed the mean accuracy of 59.9 ±0.7 % obtained from deep convolution neural network (DCNN) in recent state-of-the-art work on human activity recognition using UWB. Apart from UWB system, SPN with the data augmentation can ultimately be adopted to learn and classify time series data in various applications.

Obstructive sleep apnea (OSA) is a common sleep disorder caused by abnormal breathing. The severity of OSA can lead to many symptoms such as sudden cardiac death (SCD). Polysomnography (PSG) is a gold standard for OSA diagnosis. It records many signals from the patient's body for at least one whole night and calculates the Apnea-Hypopnea Index (AHI) which is the number of apnea or hypopnea incidences per hour. This value is then used to classify patients into OSA severity levels. However, it has many disadvantages and limitations. Consequently, we proposed a novel methodology of OSA severity classification using a Deep Learning approach. We focused on the classification between normal subjects (AHI30). The 15-second raw ECG records with apnea or hypopnea events were used with a series of one-dimensional Convolutional Neural Networks (1-D CNN) for automatic feature extraction, deep recurrent neural networks with Long Short-Term Memory (LSTM) for temporal information extraction, and fully-connected neural networks (DNN) for feature encoding from a large number of features until it closed to two classes. The main advantages of our proposed method include easier data acquisition, instantaneous OSA severity detection, and effective feature extraction without domain knowledge from expertise. To evaluate our proposed method, 545 subjects of which 364 were normal and 181 were severe OSA patients obtained from the MrOS sleep study (Visit 1) database were used with the k-fold cross-validation technique. The accuracy of 79.45% for OSA severity classification with sensitivity, specificity, and F-score was achieved. This is significantly higher than the results from the SVM classifier with RR Intervals and ECG derived respiration (EDR) signal feature extraction. The promising result shows that this proposed method is a good start for the detection of OSA severity from a single channel ECG which can be obtained from wearable devices at home and can also be applied to near real-time alerting systems such as before SCD occurs.

In this paper, the deep learning (DL) approach is applied to a joint training scheme for asynchronous motor imagery-based Brain-Computer Interface (BCI). The proposed DL approach is a cascade of one-dimensional convolutional neural networks and fully-connected neural networks (CNN-FC). The focus is mainly on three types of brain responses: non-imagery EEG (background EEG), (pure imagery) EEG, and EEG during the transitional period between background EEG and pure imagery (transitional imagery). The study of transitional imagery signals should provide greater insight into real-world scenarios. It may be inferred that pure imagery and transitional EEG are high and low power EEG imagery, respectively. Moreover, the results from the CNN-FC are compared to the conventional approach for motor imagery-BCI, namely the common spatial pattern (CSP) for feature extraction and support vector machine (SVM) for classification (CSP-SVM). Under a joint training scheme, pure and transitional imagery are treated as the same class, while background EEG is another class. Ten-fold cross-validation is used to evaluate whether the joint training scheme significantly improves the performance task of classifying pure and transitional imagery signals from background EEG. Using sparse of just a few electrode channels (Cz, C3and C4), mean accuracy reaches 71.52% and 70.27% for CNN-FC and CSP-SVM, respectively. On the other hand, mean accuracy without the joint training scheme achieve only 62.68% and 52.41% for CNN-FC and CSP-SVM, respectively.

Dramatic raising of Deep Learning (DL) approach and its capability in biomedical applications lead us to explore the advantages of using DL for sleep Apnea-Hypopnea severity classification. To reduce the complexity of clinical diagnosis using Polysomnography (PSG), which is multiple sensing platform, we incorporates our proposed DL scheme into one single Airflow (AF) sensing signal (subset of PSG). Seventeen features have been extracted from AF and then fed into Deep Neural Networks to classify in two studies. First, we proposed a binary classifications which use the cutoff indices at AHI = 5, 15 and 30 events/hour. Second, the multiple Sleep Apnea-Hypopnea Syndrome (SAHS) severity classification was proposed to classify patients into 4 groups including no SAHS, mild SAHS, moderate SAHS, and severe SAHS. For methods evaluation, we used a higher number of patients than related works to accommodate more diversity which includes 520 AF records obtained from the MrOS sleep study (Visit 2) database. We then applied the 10-fold cross-validation technique to get the accuracy, sensitivity and specificity. Moreover, we compared the results from our main classifier with other two approaches which were used in previous researches including the Support Vector Machine (SVM) and the Adaboost-Classification and Regression Trees (AB-CART). From the binary classification, our proposed method provides significantly higher performance than other two approaches with the accuracy of 83.46%, 85.39% and 92.69% in each cutoff, respectively. For the multiclass classification, it also returns a highest accuracy of all approaches with 63.70%.