ABSTRACT Multi‐channel electroencephalography (EEG) is widely recognised for its ability to non‐invasively capture dynamic brain activity with high temporal resolution. However, conventional EEG systems require meticulous setup, expert interpretation and specialised infrastructure, which limits their applicability in real‐time, resource‐constrained environments such as critical care and emergency settings. As wearable and point‐of‐care EEG technologies gain traction, there is increasing demand for low‐density, automated seizure detection methods that are clinically reliable and scalable. This study presents a data‐driven pipeline for an EEG‐based task, seizure detection, that identifies compact and clinically relevant four‐channel electrode configurations. Using standard 10–20 EEG recordings, the pipeline integrates clinical domain knowledge, preprocessing, feature extraction, and ensemble boosted classification to evaluate seizure detection performance across multiple electrode combinations. The approach was validated on both a heterogeneous local paediatric EEG dataset and an open‐source adult dataset. Our results show that several four‐channel montages achieve robust patient‐independent performance, with central and temporal electrodes frequently appearing in the top‐performing configurations. These findings support the feasibility of accurate seizure detection using low‐density EEG, even across diverse clinical populations. The primary contribution of this work is a generalisable pipeline for selecting compact EEG montages, which balances domain insight and data‐driven optimisation. This framework can be readily adapted to other EEG‐based tasks and offers a practical path toward real‐time wearable EEG monitoring in resource‐constrained environments.
ABSTRACT In this paper, we propose a photoplethysmogram (PPG) signal quality‐aware pulse rate (PR) classification and rhythm (regular/irregular) recognition framework. The proposed method comprises preprocessing, PPG signal quality assessment (PPG‐SQA), systolic peak detection and decision rules. In this study, we collected a wide variety of normal and abnormal PPG signals from standard PPG databases. Further, the PPG signals are recorded with various sensors and subjects to validate the accuracy and robustness of the method under intra‐subject, inter‐subject and sensor variability. Evaluation results demonstrate that the average peak‐to‐peak interval ()‐based PR classification method had a higher accuracy (bradycardia [BC]: 100%, normal: 99.31%, tachycardia [TC]: 95.68% for normotension [NT] blood pressure [BP] class PPG segments) as compared to the counting approach‐based PR classification method. Results further showed that the rhythm recognition method with a 4 bpm threshold outperformed, achieving a sensitivity (SE) of 99.89%, specificity (SP) of 99.33% and accuracy (ACC) of 99.80%. To reduce false alarms, we present four PPG‐SQA approaches to explore a lightweight and best PPG‐SQA with higher SE for detecting noise‐free PPG signals. Results demonstrated that the quality‐aware PR classification and rhythm recognition methods can improve the trustworthiness of cardiac arrhythmia recognition and BP analysis while reducing false alarms and computational power by discarding noisy PPG signals, which are unavoidable in continuous and long‐term health monitoring.
ABSTRACT Artificial intelligence (AI) is increasingly used in ophthalmic imaging to support early detection of vision‐threatening retinal diseases such as age‐related macular degeneration, diabetic retinopathy (DR) and choroidal neovascularisation (CNV). However, optical coherence tomography (OCT)‐based deep learning models may remain sensitive to image degradation, struggle to resolve subtle inter‐class morphological differences and show reduced reliability beyond controlled benchmark settings. We propose AESNet‐SEC, a lightweight attention‐enhanced SqueezeNet framework integrating multi‐scale dilated convolutions, spatial attention, squeeze‐and‐excitation recalibration and contrastive representation learning. The framework was evaluated on the Retinal OCT‐C8 dataset (24,000 images; eight classes) and a prospectively collected clinical dataset from Zeyaei Hospital, Iran (9732 OCT B‐scans; four classes). AESNet‐SEC achieved 98.56% accuracy on OCT‐C8 and 96.33%–96.98% across four repeated clinical evaluations. Ablation and ROC analyses showed the strongest overall discrimination for the complete configuration, while controlled Gaussian‐noise testing revealed greater degradation for DR and drusen. Grad‐CAM visualisations highlighted anatomically plausible retinal regions, with independent retinal‐specialist review providing qualitative support for the observed activation patterns. Overall, AESNet‐SEC combines strong multi‐class discrimination, computational efficiency, controlled robustness and visual interpretability, while broader patient‐disjoint, multi‐centre validation remains necessary.
ABSTRACT During a seizure, the neurons in the brain are firing abnormally. It is important to identify the symptoms to stop the spread of the virus, as a late diagnosis greatly increases the risk of infecting others. The domain is still heavily biased towards the use of just a few publicly available scalp EEG datasets (for instance, CHB/MIT) and has barely scratched the surface in relation to the use of iEEGs with little consideration for multi‐centre or private datasets. Class imbalance, where preictals outnumber interictals and icts, also presents a challenge in the training process due to the non‐stationarity, variability, and presence of noise in the EEG data. The study used the pre‐processed data and the convolution neural network model to predict the duration of the pre‐ictal phase of an epilepsy attack. To this end, the CHB‐MIT gathering dataset is used. Hand‐crafted features usually work with conventional methods. The suggested approach comprises signal metadata processing, an automatic feature extraction tool supported by signal characteristic identification and LSTM and SVM‐based result classification. The proposed model defines the pre‐ictal state considering sensitivity and specificity. Initially the detection system was configured for a two‐second window duration.
Amyotrophic lateral sclerosis (ALS) is a progressive neurodegenerative disorder affecting motor neurons, resulting in neuromuscular weakness and paralysis. Electromyography (EMG) is of vital importance for the detection of ALS. In this paper, a refined mixture of experts is proposed that automatically discriminates ALS patients from non-ALS cases using clinical EMG signals from the N2001 EMGLAB open-access dataset. The architecture consists of a 1D convolutional neural network, a temporal convolutional network and a spectrogram-based CNN to collectively learn localised temporal, long-range temporal and spectral features from EMG activity. A gating mechanism dynamically weights expert contributions and performs significantly better than equal-weight fusion. Training with focal loss and exponential moving average stabilisation addresses class imbalance and improves convergence. The proposed approach reached an AUROC of 0.9992, an F1-score of 0.9903 and a balanced accuracy of 0.9905, demonstrating strong discriminative performance and potential for real-time clinical applications.
