CN-116763253-B - Sleep apnea detection method based on double-scale convolutional neural network

CN116763253BCN 116763253 BCN116763253 BCN 116763253BCN-116763253-B

Abstract

The invention discloses a sleep apnea detection method based on a double-scale convolutional neural network, which comprises the steps of acquiring a data set of original sleep breathing; the method comprises the steps of preprocessing a data set to obtain a target fragment, extracting features of the target fragment through a double-scale convolutional neural network to obtain a target feature map, and connecting the target feature map into a full-connection classifier to classify the target feature map to obtain a sleep apnea detection result. The method for detecting the SpO 2 data by the double-scale convolutional neural network is simple to operate, convenient to carry and popularize, solves the problem of SpO 2 delay, improves the detection performance and can be widely applied to the technical field of physiological information detection.

Inventors

MA WENJUN
ZOU RUIFENG
FAN XIAOMAO
YUE HUIJUN
LEI WENBIN

Assignees

华南师范大学
中山大学附属第一医院
深圳技术大学

Dates

Publication Date: 20260505
Application Date: 20230523

Claims (8)

1. The sleep apnea detection method based on the double-scale convolutional neural network is characterized by comprising the following steps of: Acquiring a data set of original sleep breathing, including acquiring pulse oxygen saturation data to form the data set of the original sleep breathing by wearing equipment with a pulse oxygen saturation sensor; preprocessing the data set to obtain a target fragment, wherein the target fragment comprises: Performing data cleaning on the data set by using a first algorithm; The method comprises the steps of slicing a data set after data cleaning to obtain an initial segment, namely slicing the data set after data cleaning into unit segments with the length of 1 minute, marking the unit segments to obtain marked segments, synthesizing the marked segments and data 30 seconds before the next segment adjacent to the marked segments to obtain a first scale initial segment with the length of 90 seconds, taking the marked segments as intermediate segments, obtaining a third segment and a fourth segment adjacent to the marked segments, and synthesizing the third segment, the marked segments and the fourth segment to obtain a second scale initial segment; performing data equalization on the initial segment through a second algorithm and a first formula to obtain a target segment; The first algorithm is a neighbor linear interpolation algorithm, the initial segment comprises a first scale initial segment and a second scale initial segment, the target segment comprises a first scale target segment and a second scale target segment, and the first scale is smaller than the second scale; performing feature extraction on the target segment through a double-scale convolutional neural network to obtain a target feature map; And accessing the target feature map into a fully-connected classifier to classify the target feature map to obtain a sleep apnea detection result.
2. The sleep apnea detection method based on a dual-scale convolutional neural network of claim 1, wherein the step of performing data equalization on the initial segment by a second algorithm and a first formula to obtain a target segment comprises: the second algorithm is Borderline-SMOTE algorithm; The expression of the first formula is: Wherein, the For the newly generated sample to be used, For a small number of samples, Is a neighbor sample of a small number of samples, Representing a random number from 0 to 1.
3. The method for detecting sleep apnea based on the double-scale convolutional neural network according to claim 2, wherein the feature extraction of the target segment by the double-scale convolutional neural network is performed to obtain a target feature map, and the method comprises the following steps: Inputting the first scale target segment into a first feature extraction layer for feature extraction to obtain a first scale feature map; Inputting the second scale target segment into a second feature extraction layer for feature extraction to obtain a second scale feature map; and splicing the first scale feature map and the second scale feature map to obtain a target feature map.
4. The sleep apnea detection method based on the double-scale convolutional neural network according to claim 3, wherein in the step of inputting the first-scale target segment into a first feature extraction layer for feature extraction to obtain a first-scale feature map, the first feature extraction layer has 3 one-dimensional convolutional layers, namely a first convolutional layer, a second convolutional layer and a third convolutional layer; performing convolution processing through a first convolution layer, wherein the convolution kernel of the first convolution layer is 30, the number of convolution channels is 16, and the convolution step length is 1; performing convolution processing through a second convolution layer, wherein the convolution kernel of the second convolution layer is 5, the number of convolution channels is 24, and the convolution step length is 2; and carrying out convolution processing through a third convolution layer, wherein the convolution kernel of the third convolution layer is 1, the number of convolution channels is 32, and the convolution step length is 1.
5. The sleep apnea detection method based on a double-scale convolutional neural network according to claim 3, wherein in the step of inputting the second-scale target segment into a second feature extraction layer for feature extraction to obtain a second-scale feature map, the second feature extraction layer has 3 one-dimensional convolutional layers, which are a fourth convolutional layer, a fifth convolutional layer and a sixth convolutional layer respectively; performing convolution processing through a fourth convolution layer, wherein the convolution kernel of the fourth convolution layer is 30, the number of convolution channels is 16, and the convolution step length is 1; Carrying out convolution processing through a fifth convolution layer, wherein the convolution kernel of the fifth convolution layer is 5, the number of convolution channels is 24, and the convolution step length is 2; and carrying out convolution processing through a sixth convolution layer, wherein the convolution kernel of the sixth convolution layer is 1, the number of convolution channels is 32, and the convolution step length is 3.
6. The sleep apnea detection method based on a double-scale convolutional neural network according to claim 2, wherein in the step of accessing the target feature map into a fully connected classifier to classify and obtain a sleep apnea detection result, the number of classification nodes of the fully connected classifier is 64, 32 and 2.
7. Sleep apnea detection device based on double-scale convolutional neural network, characterized by comprising: The acquisition module is specifically used for acquiring pulse oxygen saturation data to form the data set of the original sleep breath by wearing equipment with a pulse oxygen saturation sensor; the preprocessing module is used for preprocessing the data set to obtain a target segment, and the preprocessing module is specifically used for: Performing data cleaning on the data set by using a first algorithm; The method comprises the steps of slicing a data set after data cleaning to obtain an initial segment, namely slicing the data set after data cleaning into unit segments with the length of 1 minute, marking the unit segments to obtain marked segments, synthesizing the marked segments and data 30 seconds before the next segment adjacent to the marked segments to obtain a first scale initial segment with the length of 90 seconds, taking the marked segments as intermediate segments, obtaining a third segment and a fourth segment adjacent to the marked segments, and synthesizing the third segment, the marked segments and the fourth segment to obtain a second scale initial segment; performing data equalization on the initial segment through a second algorithm and a first formula to obtain a target segment; The first algorithm is a neighbor linear interpolation algorithm, the initial segment comprises a first scale initial segment and a second scale initial segment, the target segment comprises a first scale target segment and a second scale target segment, and the first scale is smaller than the second scale; The feature extraction module is used for extracting features of the target segment through a double-scale convolutional neural network to obtain a target feature map; and the classification module is used for accessing the target feature map into a fully-connected classifier to classify the target feature map so as to obtain a sleep apnea detection result.
8. An electronic device comprising a processor and a memory; the memory is used for storing programs; the processor executing the program implements the method of any one of claims 1 to 6.

Description

Sleep apnea detection method based on double-scale convolutional neural network Technical Field The invention relates to the technical field of physiological information detection, in particular to a sleep apnea detection method based on a double-scale convolutional neural network. Background Sleep apnea (SLEEP APNEA, SA) refers to the symptoms of periodic apneas that occur while sleeping. The reduction of the sucked air quantity can lead to the reduction of oxygen in blood, thereby leading to sleep hypoxia, and causing long-term damage to the nervous system, the blood system and the respiratory system of a human body. SA prevalence increases rapidly with the development of obesity and aging. The existing sleep apnea detection method mainly comprises polysomnography (Polysomnography, PSG), and is still the gold standard for diagnosing SA to date, namely, a polysomnography is utilized to continuously and synchronously acquire and record a plurality of sleep physiological parameters in a sleep monitoring room, wherein the sleep physiological parameters comprise an electroencephalogram, an electrooculogram, an electromyography, an Electrocardiogram (ECG), an oronasal airflow, pulse oxygen saturation (Pulse oxygen saturation, spO 2), chest and abdomen respiratory motion, body position, snore and the like. The existing method for detecting sleep apnea mainly has the following problems: 1. Polysomnography requires that patients stay overnight in the sleep monitoring room, and strange environments often induce first night effects; polysomnography requires a subject to wear various sensors and most of the sensors are positioned on the head and the face, so that discomfort is obvious, and the sleeping is affected; 2. polysomnography is limited by sites, equipment, beds and manpower, has higher cost, increases the economic burden of patients, requires monitoring and analysis by professional medical staff in the use process and results, has high requirements on clinical experience, is difficult to popularize and apply in clinic, and has unavoidable professional and subjective errors in manual classification and judgment; 3. the existing portable sleep detection device does not consider the problem of delay of pulse oxygen saturation, namely, pulse oxygen saturation is usually generated after sleep apnea for a period of time, which can lead to inaccurate detection results. Disclosure of Invention In view of the above, the embodiment of the invention provides a simple and high-accuracy sleep apnea detection method based on a double-scale convolutional neural network. In one aspect, an embodiment of the present invention provides a sleep apnea detection method based on a dual-scale convolutional neural network, including: Acquiring a data set of original sleep breaths; preprocessing the data set to obtain a target fragment; performing feature extraction on the target segment through a double-scale convolutional neural network to obtain a target feature map; And accessing the target feature map into a fully-connected classifier to classify the target feature map to obtain a sleep apnea detection result. Optionally, the preprocessing the data set to obtain target data includes: Performing data cleaning on the data set by using a first algorithm; Carrying out data slicing on the data set subjected to data cleaning to obtain an initial segment; performing data equalization on the initial segment through a second algorithm and a first formula to obtain a target segment; the first algorithm is a neighbor linear interpolation algorithm, the initial segment comprises a first scale initial segment and a second scale initial segment, the target segment comprises a first scale target segment and a second scale target segment, and the first scale is smaller than the second scale. Optionally, the step of performing data slicing on the data set after data cleaning to obtain an initial segment includes: Cutting the data set after finishing data cleaning into unit fragments with the length of 1 minute; marking the unit fragments to obtain marked fragments; Synthesizing the marked fragment and the data of 30 seconds before the next fragment adjacent to the marked fragment to obtain a first scale initial fragment with the length of 90 seconds; And taking the marked fragment as an intermediate fragment, obtaining a third fragment and a fourth fragment adjacent to the marked fragment, and synthesizing the third fragment, the marked fragment and the fourth fragment to obtain a second-scale initial fragment. Optionally, the step of performing data equalization on the target data through the second algorithm and the first formula to obtain the target segment includes: the second algorithm is Borderline-SMOTE algorithm; The expression of the first formula is: Where x new is the newly generated sample, x minority is the minority sample, Is a neighbor sample of a few samples, rand (0, 1) represents a random number from 0 to 1. Optionally, the obt