A sleep apnea-hypopnea syndrome automatic detection and subtype classification method based on LSTM-CNN

Abstract

Sleep Apnea-Hypopnea Syndrome is a common chronic sleep respiratory disorder in adults. However, the subtype classification of SAHS and the corresponding treatment options have been a clinical challenge. In this work, we propose a novel SAHS detection and classification method, which uses C4/A1 single-channel EEG signal, oronasal flow signal and abdominal displacement signal. Firstly, we proposed a method of sleep staging based on EEG signals combined with SAHS classification, which significantly reduced the rate of false positives that appear in the waking period. Secondly, the data preprocessed by the sliding window was manipulated by Long-Short Term Memory-Convolutional Neural Network (LSTM-CNN) to identify distinct four types: normal, hypopnea events, OSAS and CSAS + MSAS. The overall classification accuracy achieves 83.94% (with 86.07% sensitivity), and the false-positive rate is 5.34% (a decrease from 28.19%), which indicates a significant improvement. Compared with PSG method, our proposed method provides promising results and improves the accuracy of SAHS detection and classification through non-invasive methods. This SAHS subtype classification method is expected to be a powerful aid to physicians in clinical diagnosis, offering the possibility of personalized treatment for patients.