EASM: An efficient AttnSleep model for sleep Apnea detection from EEG signals

This paper addresses the crucial task of automatic sleep stage classification to assist sleep experts in diagnosing sleep disorders such as sleep apnea and insomnia. The proposed solution presents a novel attention-based deep learning model called, Efficient Attention-sleep Model (EASM), designed specifically for sleep apnea detection using EEG signals. EASM incorporates a streamlined architecture that includes a modified Muti-Resolution Convolutional Neural Network (MRCNN), Adaptive Feature Recalibration (AFR), and a simplified Temporal Context Encoder (TCE) module to reduce complexity. To mitigate overfitting, ridge regression is utilized, which incorporates a penalty term to enhance model generalization. Furthermore, the proposed EASM utilizes a class-balanced focal loss function to address data imbalance issues. The effectiveness of EASM is evaluated on two publicly available datasets, SLEEP EDF-20 and SLEEP EDF-78. Comparative analysis of EASM against state-of-the-art models demonstrates its superior performance in terms of accuracy, training time, and model complexity. Notably, the proposed model achieves a 50% reduction in training time and a 55.7% decrease in complexity compared to the Attnsleep model. The EASM achieves a classification accuracy of 85.8% with minimum loss when compared to the Attnsleep model.