Application of RR series and oximetry to a statistical classifier for the detection of sleep apnoea/Hypopnoea

Abstract

In this paper we present a method for the automatic detection of sleep apnoea/Hypopnoea syndrome. This method comprises five steps. These are, signals segmentation, RR series generation, feature extraction, model training and classification. We explore the usage of the RR series and oxygen saturation (oximetry) signals both independently and jointly. Our results show that the joint usage of both improves the results obtained from the use of RR series or oximetry alone. A variety of parameterization techniques are studied in order to extract the relevant features from the signals. For the classification task we propose a two-stage strategy in which epochs are first classified by means of the power ratios. If this classification is not found reliable a Gaussian-mixture-model-based classification is applied in a second stage. A global classification of each subject is given attending to the amount of apnoea epochs. For the experiments we have used 66 subjects. The best results of our method show a 100% success in the global apnoea classification task.