EMD and PCA for the Prediction of Sleep Apnoea: A Comparative Study

Abstract

A sleep apnoea episode prediction system is presented that is based exclusively on the airflow signal. Detection of obstructive sleep apnoea (OSA) is generally carried out using polysomnography, with the data being analysed and a diagnosis formed. Being able to predict when a sleep apnoea episode is going to occur will allow for treatment to be applied before the episode becomes detrimental to the patient. Airflow signals were extracted from polysomnographic data and processed using three techniques: epoching of the flow signal, principle component analysis (PCA) and empirical mode decomposition (EMD). These processed signals were then classified using three distance functions: Euclidean, Hamming and Spearman distance. Classification of the airflow signal preceding an apnoea by Hamming distance produced the best results, with sensitivity of 81% and specificity of 76%. Reliability statistics were increase when classifying apnoea and hypopnoea episodes, with sensitivity of 95% and specificity of 100%, using Hamming distance and the empirical mode decomposition. In conclusion, classification of inspiratory airflow signal before an apnoea and hypopnoea is possible with high reliability statistics.