Automated Sleep Disorder Diagnosis Utilising EMG & EOG with Bi-LSTM Model and a Novel Reconfigurable Filter bank

Automated sleep disorders diagnosis is difficult because physiological symptoms vary considerably. Furthermore, using Polysomnography (PSG) is an expensive, time-consuming, and inconvenient procedure for the patient, where many modalities are used for a long duration. These differences make it challenging to develop efficient sleep problem detection algorithms that aid human specialists in diagnosis and therapy monitoring. As a result, we present a practical sleep disorder categorisation approach based on electromyogram (EMG) and electrooculogram (EOG) data, a simple and patient-convenient system to streamline the sleep disorders diagnostic process. Unlike many existing relevant research that employ EEG and ECG signals for feature extraction with a five-minute epoch, we adopt a pre-processing technique suited for EOG and EMG signals with a 30-s epoch. We also developed a novel, rapidly converging optimised filter bank with a constrained Equi-ripple process for frequency band separation. First, highly discriminative Hjorth parameters are computed from each subband then the extracted features are fed into the Machine Learning classifier and Bi-LSTM network to generate the model. The effectiveness of the proposed method is evaluated by classifying the sleep disorders into six classes (Insomnia, Narcolepsy, Nocturnal frontal lobe epilepsy (NFLE), Periodic leg movements (PLM), REM behaviour disorder (RBD), and healthy subjects) with maximum accuracy of 99.3%and Six stages (W, S1, S1, S3, S4, REM) with a maximum accuracy of 99.9%. The developed model is ideal for implementation in an embedded hardware device (Home based Environment) with low computing cost.