Classification of Sleep Disorder from Single Lead Non-overlapping of ECG-apnea based Non-Linear Analysis using Ensemble Approach

Abstract

The most significant determinant of quality of life is sleep quality, with better sleep resulting in a healthier and longer life. Polysomnography, or PSG, is a standardized system to get the medical records from multi-lead ECG recordings. However, PSG is a complicated, expensive and time-consuming procedure. Other alternatives include home sleep centre (HSC) development as a tool for early diagnosis and prevention of sleep disorders while keeping high accuracy. HSC uses low-cost equipment by utilizing single-lead ECG and accompanying applications. ECG is one of the media used in diagnosing and analysis of medical information related to sleep disorders. This study aims to develop a computerized sleep diagnosis application to help experts classify symptoms by investigation and evaluation of QRS morphological, time-frequency characteristics, and nonlinear analysis from single-lead ECG recordings. The classification of non-overlapping of ECG-apnea based non-linear analysis using an ensemble approach. The ensemble learning model approach, using the Boosted Tree test, yielded an accuracy of 94.7%, prediction speed of 120 obs/s and training time of 2.374 s. The QRS morphological characteristic and improved non-overlapping ECG recordings provided satisfactory diagnostic performance in sleep disorder classification for HSC usage.