Automatic detection and quantification of sleep apnea using heart rate variability

Abstract Detection of sleep apnea using electrocardiographic (ECG) parameters is noninvasive and inexpensive. Our approach is based on the hypothesis that the patient's sleep-wake cycle during episodes of sleep apnea modulates heart rate (HR) oscillations. These HR oscillations appear as low-frequency fluctuations of instantaneous HR (IHR) and can be detected using HR variability analysis in the frequency domain. The purpose of this study was to evaluate the efficacy of our ECG-based algorithm for sleep apnea detection and quantification. The algorithm first detects normal QRS complexes and R-R intervals used to derive IHR and to estimate its spectral power in several frequency ranges. A quadratic classifier, trained on the learning set, uses 2 parameters to classify the 1-minute epoch in the middle of each 6-minute window as either apneic or normal. The windows are advanced by 1-minute steps, and the classification process is repeated. As a measure of quantification, the algorithm correctly classified 84.7% of all the 1-minute epochs in the evaluation database; and as a measure of the accuracy of apnea classification, the algorithm correctly classified all 30 test recordings in the evaluation database either as apneic or normal. Our sleep apnea detection algorithm based on analysis of a single-lead ECG provides accurate apnea detection and quantification. Because of its noninvasive and low-cost nature, this algorithm has the potential for numerous applications in sleep medicine.