Pleth-SleepNet: A Unified Network for Simultaneous Sleep Stages and Disorder Classification With PPG Signal

The monitoring and diagnosis of sleep disorders have become increasingly critical in healthcare, with traditional methods often being time-consuming and reliant on expert interpretation. This project presents Pleth-SleepNet, a unified deep learning framework designed for simultaneous classification of sleep stages and disorders using plethysmography-based signals. The system leverages two advanced models: hybridCNN_GlobalPooling and LightweightAttentionCNN, which extract temporal and spatial features effectively to differentiate among four classes: healthy, mild, moderate, and severe. The models are trained and evaluated using Python with Keras, achieving classification accuracies of 0.9447 and 0.9246, respectively. By implementing an image-based approach, the framework captures complex physiological patterns, enabling robust, automated assessment of sleep quality and disorder severity. Pleth-SleepNet offers a scalable and efficient solution for clinical decision support, reducing manual intervention and enhancing early detection of sleep-related health issues.

Keywords: Sleep Disorder Classification, Sleep Stage Detection, Hybrid CNN, Attention Mechanisms, Global Pooling, Deep Learning, Plethysmography, Image Classification, Healthcare Analytics, Automated Diagnosis.