An integrated deep learning framework for EEG-based sleep disorder detection

Sleep disorders require accurate and timely detection to support effective diagnosis and intervention. However, EEG-based sleep disorder detection is challenging due to noise, signal variations, and the difficulty of extracting informative features for reliable classification. To address these limitations, this paper introduces an innovative framework called Enhanced Batch Normalization-based Parallel Convolutional Neural Network for Sleep Disorder Detection (EB-PSDD). The process begins with the acquisition of EEG signals from the CAP Sleep Database. Then the preprocessing using the Adaptive Weighted Moving Average Filtering (AWMAF) technique, which helps reduce noise and smooth the signal, ensuring its suitability for analysis. In the next phase, features are extracted using methods such as Modified Window S-Transform (MWS-T), cross-correntropy, and Common Spatial Pattern (CSP), allowing for the identification of informative characteristics within the EEG signals. For classification, a hybrid deep learning model, Hybrid LeNet-PCNN (HLNet-P), is employed. This model integrates the classic LeNet architecture with the Enhanced Batch Normalization-based Parallel Convolutional Neural Network (EBN-PCNN) to accurately analyze the extracted features and detect signs of sleep disorders. The final output indicates whether a sleep disorder is present, enabling timely detection and intervention. Experimental results show that the proposed EB-PSDD framework outperforms traditional methods, achieving a maximum accuracy of 0.949, precision of 0.799, and negative predictive value (NPV) of 0.971.