Automatic sleep breathing abnormality screening method based on BiLSTM

Abstract

The invention relates to an automatic sleep breathing abnormality screening method. Firstly, a PPG signal preprocessing method based on a Chauvenet criterion is provided to judge invalid PPG signals, secondly, a sleep breathing abnormality screening model based on BiLSTM is provided, the model is improved, so that the problems of gradient disappearance and the like can be effectively solved, and information before and after the PPG signals can be fully utilized for calculation, so that the accuracy of the model is improved.

Description

technical field [0001] The invention is a method for screening abnormal sleep breathing, in particular, an automatic screening method for abnormal sleep breathing based on BiLSTM. Background technique [0002] Abnormal sleep breathing can lead to irregular awakening of the cerebral cortex during sleep, disturbing the sleep rhythm, repeated apnea or weakening, causing hypoxemia and other diseases, and is related to cardiovascular and cerebrovascular diseases, endocrine diseases and other diseases They are closely related and seriously threaten human life, health and quality of life. Due to the expensive equipment for professional detection of sleep apnea, the detection process is cumbersome, and the analysis and processing are complicated, resulting in a low clinical diagnosis rate and untimely treatment of sleep apnea. Therefore, the research and development of portable sleep apnea monitoring equipment has attracted widespread attention at home and abroad. Among them, the m...

AI Technical Summary

Problems solved by technology

[0003] At present, the instrument for distinguishing abnormal state of sleep breathing based on pulse oximeter is relatively mature, but the processing of pulse oximeter signal (PPG) and the establishment of subsequent classification models will directly affect the final result of this method for disease discrimination, so this method is The bottleneck of popularization and application is also the focus and hotspot of research at home and abroad

The PPG signal will be interfered by various factors during the acquisition process, especially the invalid signal generated by spike noise and poor contact of the sensor, which seriously affects the accuracy of the subsequent detection model

With the development of deep learning technology, the classification and detection of signals tends to be processed by deep learning methods. Representative models include convolutional neural network (CNN) and recurrent neural network (RNN). However, these deep models have Problems such as gradient disappearance affect the accuracy of model classification check

Images