Data Segmentation Effects on Oxygen in Blood Sleep Apnea Detection CNN-based Model

This work aims to study the effects of data segmentation and data labeling for predicting sleep apnea events with SpO2 data using a one-dimension CNN-based model. Three workloads are used in this study. Sensitivity, specificity and accuracy are used as the measurements. The results show that the 90 second-interval data segmentation provides the best performances on two workloads including the sleep disorders service and research center at Songklanakarin hospital, Thailand workload and the St. Vincent's University Hospital dataset. On the Apnea-ECG dataset, however, the 90 second-interval data segmentation provides the best sensitivity value while the specificity and accuracy values are at the competitive level with previously proposed methods.