Automatic recognition of obstructive sleep apnoea syndrome using power spectral analysis of electrocardiogram and Hidden Markov Models

Obstructive sleep apnoea syndrome (OSA) is a very common disorder in breathing during sleep. OSA is considered as clinically relevant when the breath stops during more than 10 seconds and occurs more than five times per sleep hour. In this work, we investigate a noninvasive automatic approach to classify sleep apnoea events based on power spectral analysis for the feature extraction of the ECG records and hidden Markov models (HMMs). Based on Bayesian inference criterion (BIC), the proposed HMM training algorithm is able to select the optimal number of states corresponding to each set of training features. For every state number, each iteration is initialized by the most appropriate model using data clustering, and by the rejection of the least probable state of the previous iteration. Both off-line and on-line schemes have been proposed. Only electrocardiogram (ECG) records are considered for the detection of OSA. In this preliminary work, we report training procedures and validation results of the models on whole night digitized ECG signals recorded from 70 subjects with normal and OSA breathing events obtained from the physionet database.