Sleep Apnea Identification Through Vertical Respiratory Movement Using Region of Interest and Recurrent Neural Networks

Sleep Apnea is a sleep state of a person who has breathing problems where the patient often stops breathing during sleep. Sleep Apnea can be identified using Polysomnography. Nevertheless, the equipment that touches the patient produces so more difficulty sleeping. Monitoring Sleep Apnea through respiratory movement video is beneficial even though video processing during sleeping will take the problem in memory. It involves all frames and the relation between frames. The central is vertical respiration movement, so processing uses Region of Interest (ROI) to focus on that area. Segmentation of the head and chest area focuses on the vertical respiratory movement in each video frame on the Sleep Apnea monitor. This paper proposed a method of identifying Sleep Apnea using ROI and Recurrent Neural Networks (RNN). The experiment comparing chest area processing alone was more accurate than head and chest area using the Gated Recurrent Unit (GRU) architecture, 94.44%. The head area gave an accuracy of 83.33%. Both areas made it better using the LSTM architecture with an accuracy of 73.10% than the GRU with an accuracy of 68.10%, considering the LSTM processes mixed data. GRU provided better accuracy when used for a single region.