Electrocardiogram based Sleep Apnea Detection using SVM Classifier

Sleep apnea is a common disorder that affects breathing during sleep, and it can lead to serioushealth problems if left untreated. In this study, we propose a machine learning approach usingelectrocardiogram data to detect sleep apnea.Background and aims: Sleep apnea is a common sleep disorder characterized by the cessationof breathing during sleep. It can cause daytime fatigue, headaches, and even serious health problemssuch as high blood pressure, stroke, and heart disease if left untreated. Polysomnography isthe gold standard for sleep apnea diagnosis, but it is time-consuming and expensive. Therefore,developing a reliable and efficient method for sleep apnea detection is crucial.Method: We used electrocardiogram data from the PhysioNet Apnea-ECG database to detectsleep apnea. The database consists of ۷۰ records, with recordings varying in length from slightlyless than ۷ hours to nearly ۱۰ hours. The data is divided into ۱-minute segments, and features areextracted from each segment. Feature extraction is done in two phases. In the first phase, differenttime and frequency domain features are extracted from the raw data. Some of these featuresinclude skewness, kurtosis, mean, standard deviation, power spectrum analysis, and entropy. Inthe second phase, electrocardiogram data is processed to obtain the RR intervals and R-peakamplitudes. The RR interval is defined as the time interval between two consecutive R peaks.To locate R peaks, the Hamilton algorithm is utilized. The R peaks are subsequently used forthe computation of RR intervals. Then, these new signals are analyzed, and different features areextracted, such as MRR, MHR, RMSSD, pNN۵۰, VLF, and HF. In the final phase, all features areconcatenated and fed to different traditional classifiers using a ۵-fold cross-validation technique.Finally, the best performance was achieved by utilizing SVM.Results: Our proposed method achieved an accuracy of ۸۵.۶۳%, a sensitivity of ۷۹.۰۳%, a specificityof ۸۹.۷۱%, and an F-score of ۸۰.۷۹%. Our method outperformed state-of-the-art methodsthat use traditional machine learning methods, with reported accuracies of ۸۰.۷%, ۸۲.۱۲%, and۷۹.۳۹%, respectively.Conclusion: In conclusion, our study demonstrates the effectiveness of using machine learningand electrocardiogram data to detect sleep apnea. While deep learning methods have shownpromising results in this field, they are often computationally expensive. In contrast, our proposedmethod is computationally lightweight and faster. We achieved promising results using a combinationof time and frequency domain features and an SVM classifier. Previous studies only usedR peak and RR interval features. In this study, we showed that by combining additional featuresextracted from raw data, performance significantly improved. For future work, we plan to furtherimprove our feature extraction procedure and analyze features more accurately. Specifically, wewill explore the use of additional features from the electrocardiogram signals. Overall, our proposedmethod holds great potential for improving the diagnosis and treatment of sleep apnea.

کلیدواژه ها:

Sleep apnea ، electrocardiogram ، SVM ، feature extraction