A Sequence-to-Sequence Sleep Disorder Detection Method Based on Fully Convolutional Networks

Abstract

The invention discloses a sequence-to-sequence sleep disorder detection method based on a full convolutional network. The method acquires a data set that needs to detect sleep disorders, preprocesses the data set, and constructs a full convolutional network and a long-short-term memory network. Sleep wake detection network model and model training, using the trained sleep wake detection network model for sleep disorder detection. The present invention reduces the workload and improves the processing efficiency by uniformly preprocessing the multimodal biological signals, and retains the context information of the data to improve the detection accuracy; the present invention directly provides The sleep-wake index realizes the automation of sleep disorder detection, and through the sequence-to-sequence detection method, it retains the correlation of data in time series and achieves better detection results.

Description

technical field [0001] The invention belongs to the technical field of sleep detection, and in particular relates to a sequence-to-sequence sleep disorder detection method based on a fully convolutional network. Background technique [0002] Sleep quality is closely related to human health. Sleep disorders can affect people's mental state, which can lead to health problems, especially obesity, depression, mental illness (such as neurasthenia), and some cardiovascular and cerebrovascular diseases that have a higher risk of death. Surveys show that people with sleep disorders have a 2.6 times higher risk of myocardial infarction than ordinary people, and a 1.5-4 times higher risk of stroke. [0003] Breathing disorders during sleep are the main cause of poor sleep quality. Sleep disturbances can have many causes, such as spontaneous arousals, arousals associated with respiratory effort, teeth grinding, snoring, hypopnea, apnea, periodic leg movements, tidal breathing, or part...

AI Technical Summary

Problems solved by technology

There are three main problems with this type of method: first, the preprocessing part of the data is very complicated and cumbersome

Most of the original signals have a certain amount of noise, and different signal preprocessing methods need to be used to remove the noise for different biological signals, and then extract the corresponding features; secondly, according to the definition of the American Association of Sleep Medicine, the automatic detection of sleep disorders means that at least one EEG signal and an electromyography (EMG) signal are researched and deduced. However, it is difficult to analyze events in different biological signals at the same time. Therefore, most of the research is done on a single signal, and a single The signal of the signal cannot identify the wake-up area very accurately; the third point is that most of the methods have sliced ​​the original data. Since the duration of sleep and awakening accounts for a small proportion (<5%) of the total sleep duration, so for To ensure data balance, most of the data needs to be discarded

Since sleep arousal is defined as a sudden change in the frequency of the EEG signal, lasting at least 3 seconds, and preceded by at least 10 seconds of stable sleep, the detection of sleep disorders needs to consider the temporal correlation of the data, so data slicing will lead to Some context information is lost, which affects the accuracy

Images