0270 Obstructive Sleep Apnea detection using ECG morphology and Machine Learning

Abstract Introduction Obstructive Sleep Apnea (OSA) is characterized by reduction in airflow. Hypopnea a smaller airflow compared to apnea but also accompanied drop oxygen saturation leading sympathetic activation. Frequent OSA events are correlated with incidence of cardiovascular diseases. Measuring changes requires overnight polysomnography that expensive. Electrocardiogram (ECG) available through wearable devices at home and therefore, detecting using ECG can make diagnostics more accessible. This paper studies the use morphology an ensemble machine learning algorithm detect events. The designed light weight ensure it be implemented on embedded processor. Methods data from Apnea, Bariatric surgery, CPAP (ABC) study provided National Research Resource used for analysis. Twenty-six subjects diagnosed severe considered single night before treatment. Since have minimum 10s duration, non-overlapping windows labels technician scoring used. Features extracted PQRST complex majority voting across multiple algorithms select top 7 features (SDNN, RMSSD, P-P interval, P-duration, P-R T-duration T-P interval) highest explanatory power. We train random forest classifier leave-one-out methodology where 25 training 26th subject testing (all permutations used). sensitivity, precision F1 score evaluation. Results detects precise occurrence (onset offset) as well total number during night. precision, sensitivity scores 70%, 96% 80% respectively. prediction leans towards higher suffer OSA. uncover 32% false positives significant SPO2 desaturation pointing inaccuracy manual scoring. Assuming these correct predictions increases 79%. Conclusion classification demonstrates feasibility wearables measure setting. Support (if any)