Hybrid Convolutional Neural Network - Long Short-Term Memory Model for Automated Detection of Sleep Stages

Sleep is essential to maintain the physical and mental fitness of a living being. An irregular sleep cycle may cause extreme drowsiness, snoring and sleep interruptions, leading to sleep disorders. It is identified using a diagnostic tool called Polysomnography (PSG) to conduct a sleep study, which is manually scored by the physician in the sleep lab. It is considered a complex task for the physician to manually analyze the sleep stages based on PSG scores. This work proposes an Intelligent Model based on Deep Learning algorithms to diagnose sleep disorders through sleep stage classification. Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and a Hybrid Model (CNN-LSTM) are employed to characterize sleep stages. The developed model categorizes the patient's sleep patterns using PSG signals, by which the doctor can analyze the sleep stages. This model assists clinicians to analyze sleep patterns to diagnose sleep stages with minimum effort. The comparative analysis of this study demonstrates that the Hybrid Model algorithm gives better accuracy than CNN and LSTM models.