An approach to the enhancement of sleep apnea detection by means of detrended fluctuation analysis of RR intervals

In this paper, Detrended Fluctuation Analysis (DFA) of Heart Rate Variability (HRV) is applied in order to study the performance of a classification system of Obstructive Sleep Apnea (OSA), that integrates other variables as cepstrum coefficients and filter banks (FBANK) obtained from HRV. The database contains 70 records, divided into two equal-sized sets: a learning set and a test set. Each recording includes a continuous digitized single channel ECG signal and a set of apnea annotations, where a human expert classifies each minute indicating normal breathing or OSA, on the basis of a complete polysomnography (PSG). An automatic statistical classification method based on QDA (Quadratic Discriminant Analysis) and Logistic Regression (LR) is applied to the classification of sleep apnea epochs. Particularly QDA presents an accuracy of 82.4% (auc=0.9) when FBANK and cepstrum coefficients are applied. The performance increases to 84.3% (auc=0.91) when DFA is added to the model. Similar improvement with LR when DF A is added can be reached 81.5% (auc=0.88) vs 84.2% (auc=0.91).