An Effective Method for Distinguishing Sleep Apnea and Hypopnea Based on ECG Signals

Good sleep quality is essential to our life and work. The two major challenges in evaluating sleep quality are the scoring of sleep stages and sleep apnea detection. In addition to sleep apnea, hypopnea is also a respiratory event that occurs at an equal or even higher frequency compared to that of apnea. However, the two phenomena are not only different in terms of physical manifestation, there are also significant differences in the methods and means of clinical treatment. Therefore, in this study we propose an effective algorithm to distinguish between hypopnea and apnea events. We consider two aspects of feature extraction. On the one hand, from the original time-domain waveform of electrocardiogram signals, we extract the time difference sequence between the appearance of the R-R wave in two adjacent QRS waves (RR intervals) and explore the way they change when apnea and hypopnea event occur. On the other hand, considering that the degree of sequence disorder caused by hypopnea and apnea events is different, multi-scale entropy and complexity are chosen as the features that describe this change. The results obtained using the proposed algorithm show that the classification between hypopnea events and any type of apnea events all reach an accuracy rate greater than 90%. For obstructive apnea events, which occur most frequently among the three different types of apneas, the classification accuracy, sensitivity, specificity and F1-score were 91.78%, 92.21%, 91.30%, 92.20%, respectively. The classification results between central apnea events and hypopnea events reached an accuracy of 93.18%, sensitivity of 95.00%, specificity of 91.67% and F1-score of 92.68%. At the same time, the algorithm achieved an accuracy of 92.59%, sensitivity of 90.38%, specificity of 94.64% and F1-score of 92.16% when classifying mixed apnea and hypopnea events. In the classification of normal breathing, hypopnea events and apnea events, the total accuracy rate achieved was 91.92%, with that for normal breathing reaching 93.62%, for hypopnea events reaching 90.39%, and 91.91% for apnea events, with a Kappa coefficient of 0.99. The algorithm proposed in this study constitutes an efficient and convenient approach for apnea and hypopnea event detection and classification.