Sleep Apnea Detection Using Wireless Wearable Single-Lead ECG: A One Dimensional Convolutional Neural Network Approach

Background: Polysomnography, the standard diagnostic tool for sleep apnea, is known for its high cost and time consumption.It requires multiple sensors and a hospital visit, and the unfamiliar environment can affect the sleep test results.Additionally, labeling sleep apnea is a labor-intensive process that requires a lot of analysis time from the specialist.To address these issues, this paper explores the potential of wireless wearable electrocardiogram (ECG) devices to enhance convenience and enable remote monitoring and diagnosis of sleep apnea. Methods:The study proposed a one-dimensional convolutional neural network (1-D CNN) model that uses ECG and respiratory signals obtained from patch-type wearable devices to detect sleep apnea.The obtained signals were preprocessed through segmentation and normalization and proposed as input to the model.A class weighting technique was introduced to address data imbalance. Results:The developed 1-D CNN model's performance was evaluated to determine the most reliable signals for the development of a sleep apnea detection system.The respiration model based on the 1-D CNN achieved an average accuracy of 88.37%, a F1 score of 60.67% and ECG model achieved averaged accuracy of 95.38, F1 score of 51.46% across 4-fold cross-validation.These results suggest the possibility that there may be mutual compensation between ECG and respiration signals. Conclusion:This approach, combining the convenience of wearable devices with the analytical power of deep learning, can effectively detect and diagnose sleep apnea without requiring a hospital visit, thereby reducing the diagnostic time.This study confirms the potential of automating sleep apnea classification using wearable ECG devices and 1-D CNN, improving diagnostic convenience.