Automated Classification of Sleep Apnea and Hypopnea on Polysomnography Data

Abstract

Polysomnography (PSG) is a sleep study where multiple parameters of a subject are continuously monitored to detect sleep disorders that can have adverse health effects. This study uses Machine Learning techniques to automatically detect and classify apneas and hypopneas in PSG data. By incorporating features that sleep analysts look for in PSG data we present a machine learning based approach to automatically detect the presence of hypopnea or apnea and classify the type of pathology. This paper presents a visualization of the PSG study, explicates the design of features and provides a comparison between the approaches of learning from the computed features versus the signal directly. Our results demonstrate that a hierarchical SVM model trained on a small set of features yields an accuracy of 82.6% with a high precision of 86%.