1. A method for detecting obstructive sleep apnea based on deep learning is characterized by comprising the following steps:

inputting physiological data of a detected object into a pre-trained physiological data obstructive sleep apnea detection model, and outputting a corresponding Apnea Hypopnea Index (AHI) classification result;

inputting image data of a detected object into a pre-trained image data obstructive sleep apnea detection model, and outputting a corresponding AHI classification result, wherein the image data comprises: shooting a plurality of head images of the head area of the detected object from a plurality of preset angles, wherein the head image of each preset angle is input into different image data obstructive sleep apnea detection models;

and inputting the AHI classification result output by the physiological data obstructive sleep apnea detection model and the AHI classification result output by each image data obstructive sleep apnea detection model into an integrated learning algorithm model, and outputting the AHI classification result of the detected object.

2. The method of claim 1, wherein the method further comprises:

obtaining sample physiological data, wherein the sample physiological data comprises: physiological data of sample subjects with different AHI tags;

dividing the sample physiological data into training physiological data, verification physiological data and test physiological data according to a preset proportion;

according to the training physiological data, any one of the following network models is trained by using a deep learning method to obtain the physiological data obstructive sleep apnea detection model: the system comprises an Xgboost model, a Light GBM model, a neural network model, a support vector machine model, a decision tree classification model and a Bayesian decision classification model;

verifying the physiological data obstructive sleep apnea detection model obtained by training according to the verified physiological data until the model accuracy rate meets the preset condition;

and testing the physiological data obstructive sleep apnea detection model which meets the preset conditions according to the test physiological data.

3. The method of claim 1, wherein the method further comprises:

acquiring sample image data, wherein the sample image data comprises: shooting head areas of sample objects with different AHI labels from a plurality of preset angles to obtain a plurality of head images;

dividing the sample image data into training image data, verification image data and test image data according to a preset proportion;

according to the training image data, training any one of the following network models by using a deep learning method to obtain a plurality of image data obstructive sleep apnea detection models corresponding to a plurality of preset angle head images: the system comprises an Xgboost model, a Light GBM model, a neural network model, a support vector machine model, a decision tree classification model and a Bayesian decision classification model;

verifying each image data obstructive sleep apnea detection model obtained by training according to the verification image data until the model accuracy rate meets the preset condition;

and testing the image data obstructive sleep apnea detection model which meets the preset conditions according to the test image data.

4. The method of claim 1, wherein inputting image data of a detected subject into a pre-trained image data obstructive sleep apnea detection model and outputting corresponding AHI classification results comprises:

inputting image data of a detected object into a pre-trained face recognition model, and outputting corresponding face image data, wherein the face recognition model is obtained by training a neural network model;

and inputting the face image data output by the face recognition model into a pre-trained image data obstructive sleep apnea detection model, and outputting a corresponding AHI classification result.

5. The method of claim 4, wherein before inputting the image data of the detected object into the pre-trained face recognition model and outputting the corresponding face image data, the method further comprises:

acquiring sample image data, wherein the sample image data comprises: shooting head areas of sample objects with different AHI labels from a plurality of preset angles to obtain a plurality of head images;

performing cluster analysis on head images with different preset angles in sample image data to obtain a high-level representation subset belonging to each type of OSA;

training the following neural network model according to the high-level representation subset belonging to each type of OSA to obtain a face recognition model belonging to each type of OSA: a residual error neural network model and a convolution neural network model based on an inverted residual error module.

6. The method of claim 1, wherein the AHI classification result is a probability of belonging to a different AHI classification, and the integrated learning algorithm model is configured to perform a weighted summation of the probability of belonging to the different AHI classification of the detected subject outputted by the physiological data obstructive sleep apnea detection model and the probability of belonging to the different AHI classification of the detected subject outputted by each image data obstructive sleep apnea detection model to obtain the AHI classification result of the detected subject.

7. The method according to any one of claims 1 to 6, wherein the preset angle comprises at least five directions: the front surface and the side surface face to the left, the side surface faces to the right, 45 degrees face to the left and 45 degrees face to the right.

8. An obstructive sleep apnea detecting apparatus based on deep learning, comprising:

the physiological data processing module is used for inputting the physiological data of the detected object into a pre-trained physiological data obstructive sleep apnea detection model and outputting a corresponding Apnea Hypopnea Index (AHI) classification result;

the image data processing module is used for inputting image data of a detected object into a pre-trained image data obstructive sleep apnea detection model and outputting a corresponding AHI classification result, wherein the image data comprises: shooting a plurality of head images of the head area of the detected object from a plurality of preset angles, wherein the head image of each preset angle is input into different image data obstructive sleep apnea detection models;

and the obstructive sleep apnea detecting module is used for inputting the AHI classification result output by the physiological data obstructive sleep apnea detecting model and the AHI classification result output by each image data obstructive sleep apnea detecting model into the integrated learning algorithm model and outputting the AHI classification result of the detected object.

9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the computer program implements the deep learning based obstructive sleep apnea detection method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program for executing the deep learning based obstructive sleep apnea detecting method of any one of claims 1 to 7.

Background

This section is intended to provide a background or context to the embodiments of the invention that are recited in the claims. The description herein is not admitted to be prior art by inclusion in this section.

Obstructive Sleep Apnea Hypopnea Syndrome (OSAHS, OSA for short) is one of common Sleep respiratory disorders, mainly caused by repeated collapse of the upper respiratory tract during Sleep, repeated stop of the mouth-nose respiratory airflow, and accompanied by cortex arousal or blood oxygen saturation reduction, so as to cause repeated intermittent hypoxia and hypercapnia, and mainly clinically manifested by snoring during night Sleep, daytime sleepiness, morning dry mouth, headache and the like. Repeated episodes of partial or complete collapse of the upper airway during sleep lead to a sustained reduction in airflow (hypopnea) or absence (apnea) for at least 10 seconds with concomitant cortical arousal or reduced blood oxygen saturation. OSA is a major cause of excessive drowsiness, decreased quality of life, impaired performance, and increased risk of motor vehicle collisions. OSA is associated with an increased incidence of hypertension, type 2 diabetes, atrial fibrillation, heart failure, coronary heart disease, stroke, and death.

Generally, the presence and severity of OSA is quantified by the Apnoea Hypopnea Index (AHI), which is defined as the number of apneas plus hypopneas per hour of sleep time (or per hour recorded by home testing). When the apnea and hypopnea repeatedly attacks more than 30 times or the apnea hypopnea index AHI is more than or equal to 5 times/h within 7 hours of sleep every night, the patient is considered to be OSA.

Statistically, 17.4% of women and 33.9% of men in the united states have at least mild OSA, defined as AHI with 5 to 14.9 episodes per hour of sleep, while 5.6% of women and 13.0% of men have moderate (AHI of 15-29.9) or severe (AHI 30) OSA. In 2018, about 1.76 hundred million people exist in mild OSA patients (AHI is more than or equal to 5) in China, and about 6600 million people exist in more than moderate OSA patients (AHI is more than or equal to 15). The patient with mild OSA can generally relieve the symptoms by improving living habits such as body building, weight losing, smoking cessation, sleeping posture changing and the like, the patient is not easy to find, the breathing machine is needed only by the patient with moderate OSA, the permeability is only calculated by 10%, 660 ten thousand breathing machines are still needed in China in the OSA patients, the annual sales of 16-17 ten thousand breathing machines in China in 2018 are still needed, and the efficiency is still needed to be improved in the aspect of diagnosing the OSA.

The prior art diagnoses whether the patient suffers from OSA based on polysomnography (such as neurophysiologic, cardiac and respiratory signals), which not only needs the patient to wear various PSG devices, but also monitors the sleep record of the patient all night, and only if the patient has corresponding symptoms, the patient is considered to suffer from OSA, and the sleep apnea or hypopnea without any symptoms, which is found in the sleep record, is often not considered to be OSA, so that the patient is difficult to know the condition of the patient in time without early treatment.

Therefore, how to provide a convenient and accurate method for detecting obstructive sleep apnea is a technical problem to be solved urgently at present.

Disclosure of Invention

The embodiment of the invention provides a method for detecting obstructive sleep apnea based on deep learning, which is used for solving the technical problem that the existing method for diagnosing OSA based on the night sleep record of a patient is complex in operation, and comprises the following steps: inputting physiological data of a detected object into a pre-trained physiological data obstructive sleep apnea detection model, and outputting a corresponding Apnea Hypopnea Index (AHI) classification result; inputting image data of a detected object into a pre-trained image data obstructive sleep apnea detection model, and outputting a corresponding AHI classification result, wherein the image data comprises: shooting a plurality of head images of the head area of the detected object from a plurality of preset angles, wherein the head image of each preset angle is input into different image data obstructive sleep apnea detection models; and inputting the AHI classification result output by the physiological data obstructive sleep apnea detection model and the AHI classification result output by each image data obstructive sleep apnea detection model into an integrated learning algorithm model, and outputting the AHI classification result of the detected object.

The embodiment of the invention also provides a device for detecting obstructive sleep apnea based on deep learning, which is used for solving the technical problem that the operation of the existing method for diagnosing OSA based on the night sleep record of a patient is complex, and the device comprises: the physiological data processing module is used for inputting the physiological data of the detected object into a pre-trained physiological data obstructive sleep apnea detection model and outputting a corresponding Apnea Hypopnea Index (AHI) classification result; the image data processing module is used for inputting image data of a detected object into a pre-trained image data obstructive sleep apnea detection model and outputting a corresponding AHI classification result, wherein the image data comprises: shooting a plurality of head images of the head area of the detected object from a plurality of preset angles, wherein the head image of each preset angle is input into different image data obstructive sleep apnea detection models; and the obstructive sleep apnea detecting module is used for inputting the AHI classification result output by the physiological data obstructive sleep apnea detecting model and the AHI classification result output by each image data obstructive sleep apnea detecting model into the integrated learning algorithm model and outputting the AHI classification result of the detected object.

The embodiment of the invention also provides a computer device, which is used for solving the technical problem of complex operation of the existing method for diagnosing OSA based on the sleep record of a patient all night.

The embodiment of the invention also provides a computer readable storage medium for solving the technical problem of complex operation of the existing method for diagnosing OSA based on the sleep record of the patient all night, wherein the computer readable storage medium stores a computer program for executing the method for detecting the obstructive sleep apnea based on the deep learning.

In the embodiment of the invention, after acquiring the physiological data of the detected object and a plurality of head images obtained by shooting the head area of the detected object from a plurality of angles, inputting the physiological data of the detected object into a pre-trained physiological data obstructive sleep apnea detection model, outputting a corresponding apnea hypopnea index AHI classification result, and a plurality of head images of the detected object are input into pre-trained image data obstructive sleep apnea detection models corresponding to different preset angles, and outputting a corresponding AHI classification result, and finally inputting the AHI classification result output by the physiological data obstructive sleep apnea detection model and the AHI classification result output by each image data obstructive sleep apnea detection model into an integrated learning algorithm model and outputting the AHI classification result of the detected object.

Compared with the technical scheme of diagnosing OSA based on the night sleep record of the patient in the prior art, the obstructive sleep apnea detecting method provided by the embodiment of the invention only needs to collect the head images of the patient at multiple angles, does not need the user to wear various PSG devices, is simple to operate, and can detect the obstructive sleep apnea condition with a low AHI value, so that the user can treat OSA in time at the early stage.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts. In the drawings:

fig. 1 is a flowchart of a method for detecting obstructive sleep apnea based on deep learning according to an embodiment of the present invention;

fig. 2 is a flowchart of an implementation of a method for detecting obstructive sleep apnea based on deep learning according to an embodiment of the present invention;

FIG. 3 is a flow chart of model training provided in an embodiment of the present invention;

FIG. 4 is a flowchart illustrating an exemplary training process of an obstructive sleep apnea detection model according to an embodiment of the present disclosure;

FIG. 5 is a flowchart of a physiological data obstructive sleep apnea detection model training process provided in an embodiment of the present invention;

fig. 6 is a schematic diagram of an obstructive sleep apnea detecting apparatus based on deep learning according to an embodiment of the present invention;

fig. 7 is a schematic diagram of a computer device provided in an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. The exemplary embodiments and descriptions of the present invention are provided to explain the present invention, but not to limit the present invention.

Deep learning is a sub-field of Artificial Intelligence (AI), and is rapidly developing in the medical field. The deep learning system may accept a variety of data types, such as images or time series, that are to be processed through multiple processing layers based on a machine learning model to learn a representation of the data step by step and ultimately provide an output. In the healthcare field, deep learning techniques may have potential uses for diagnosis, treatment, population health management, and management and regulation.

An embodiment of the present invention provides a method for detecting obstructive sleep apnea based on deep learning, and fig. 1 is a flowchart of the method for detecting obstructive sleep apnea based on deep learning, as shown in fig. 1, the method includes the following steps:

s101, inputting the physiological data of the detected object into a pre-trained physiological data obstructive sleep apnea detection model, and outputting a corresponding Apnea Hypopnea Index (AHI) classification result.

Optionally, in the method for detecting obstructive sleep apnea based on deep learning provided in the embodiment of the present invention, when training a physiological data obstructive sleep apnea detection model, an adopted network model includes any one of the following: the system comprises an Xgboost model, a Light GBM model, a neural network model, a support vector machine model, a decision tree classification model and a Bayesian decision classification model.

S102, inputting image data of a detected object into a pre-trained image data obstructive sleep apnea detection model, and outputting a corresponding AHI classification result, wherein the image data comprises: the method comprises the steps of shooting a plurality of head images of a head area of a detected object from a plurality of preset angles, wherein the head images of each preset angle are input into different image data obstructive sleep apnea detection models.

It should be noted that, in order to simplify the model operation and obtain a better classification result, the preset angles adopted in the embodiment of the present invention include, but are not limited to, the following five directions: the front surface and the side surface face to the left, the side surface faces to the right, 45 degrees face to the left and 45 degrees face to the right.

Optionally, in the method for detecting obstructive sleep apnea based on deep learning provided in the embodiment of the present invention, when training the image data obstructive sleep apnea detection model at each preset angle, the network model used includes any one of the following: the system comprises an Xgboost model, a Light GBM model, a neural network model, a support vector machine model, a decision tree classification model and a Bayesian decision classification model.

S103, inputting the AHI classification result output by the physiological data obstructive sleep apnea detection model and the AHI classification result output by each image data obstructive sleep apnea detection model into the integrated learning algorithm model, and outputting the AHI classification result of the detected object.

It should be noted that the number of the image data obstructive sleep apnea detection models adopted in the embodiment of the present invention depends on the shooting angle, if the shooting angle is N directions, N corresponding image data obstructive sleep apnea detection models need to be trained, and finally, the classification results output by the N image data obstructive sleep apnea detection models and the classification results output by the 1 physiological data obstructive sleep apnea detection model are combined to obtain the final classification result.

In an embodiment, in the method for detecting obstructive sleep apnea based on deep learning provided in the embodiment of the present invention, the AHI classification results output by the physiological data obstructive sleep apnea detection model and the image data obstructive sleep apnea detection models are probabilities belonging to different AHI classifications, and the integrated learning algorithm model is configured to perform weighted summation on the probabilities that the detected object output by the physiological data obstructive sleep apnea detection model belongs to different AHI classifications and the probabilities that the detected object output by the image data obstructive sleep apnea detection models belongs to different AHI classifications, so as to obtain the AHI classification result of the detected object.

In an embodiment, before inputting physiological data of a detected subject into a pre-trained physiological data obstructive sleep apnea detection model and outputting a corresponding apnea hypopnea index AHI classification result, the method for detecting obstructive sleep apnea based on deep learning provided in an embodiment of the present invention may further include the following steps: obtaining sample physiological data, wherein the sample physiological data comprises: physiological data of sample subjects with different AHI tags; dividing sample physiological data into training physiological data, verification physiological data and test physiological data according to a preset proportion; according to the training physiological data, training any one of the following network models by using a deep learning method to obtain a physiological data obstructive sleep apnea detection model: the system comprises an Xgboost model, a Light GBM model, a neural network model, a support vector machine model, a decision tree classification model and a Bayesian decision classification model; verifying the physiological data obstructive sleep apnea detection model obtained by training according to the verified physiological data until the model accuracy rate meets the preset condition; and testing the physiological data obstructive sleep apnea detection model which meets the preset conditions according to the tested physiological data.

Further, before inputting the image data of the detected object into the pre-trained image data obstructive sleep apnea detection model and outputting the corresponding AHI classification result, the method for detecting obstructive sleep apnea based on deep learning provided in the embodiment of the present invention may further include the following steps: acquiring sample image data, wherein the sample image data comprises: shooting head areas of sample objects with different AHI labels from a plurality of preset angles to obtain a plurality of head images; dividing sample image data into training image data, verification image data and test image data according to a preset proportion; according to the training image data, training any one of the following network models by using a deep learning method to obtain a plurality of image data obstructive sleep apnea detection models corresponding to a plurality of preset angle head images: the system comprises an Xgboost model, a Light GBM model, a neural network model, a support vector machine model, a decision tree classification model and a Bayesian decision classification model; verifying each image data obstructive sleep apnea detection model obtained by training according to verification image data until the model accuracy rate meets a preset condition; and testing the image data obstructive sleep apnea detection model which meets the preset conditions according to the test image data.

Further, the method for detecting obstructive sleep apnea based on deep learning provided in the embodiment of the present invention may be implemented by the following steps when inputting image data of a detected object into a pre-trained image data obstructive sleep apnea detection model and outputting a corresponding AHI classification result: inputting image data of a detected object into a pre-trained face recognition model, and outputting corresponding face image data, wherein the face recognition model is obtained by training a neural network model; and inputting the face image data output by the face recognition model into a pre-trained image data obstructive sleep apnea detection model, and outputting a corresponding AHI classification result.

In one embodiment, before inputting image data of a detected object into a pre-trained face recognition model and outputting corresponding face image data, the method for detecting obstructive sleep apnea based on deep learning provided in the embodiments of the present invention further includes the following steps: acquiring sample image data, wherein the sample image data comprises: shooting head areas of sample objects with different AHI labels from a plurality of preset angles to obtain a plurality of head images; performing cluster analysis on head images with different preset angles in sample image data to obtain a high-level representation subset belonging to each type of OSA; training the following neural network model according to the high-level representation subset belonging to each type of OSA to obtain a face recognition model belonging to each type of OSA: a residual error neural network model and a convolution neural network model based on an inverted residual error module.

Fig. 2 is a flowchart of an implementation of a method for detecting obstructive sleep apnea based on deep learning according to an embodiment of the present invention, and as shown in fig. 2, the method specifically includes the following steps:

s201, image data and physiological data of the detected object are acquired.

In specific implementation, the acquired image data includes, but is not limited to, a head image with five angles of front, side facing left, side facing right, 45 degrees facing left, and 45 degrees facing right of the detected object.

S202, image preprocessing is performed on the input of the obtained image data, only the face image is retained, and the image is scaled to the same size.

In the implementation, known data and corresponding AHI tags are used as training samples, wherein the AHI tags corresponding to the known data may be determined by a corresponding patient after wearing the PSG device for monitoring overnight, which is not limited by the invention. Using an AHI tag for OSA classification, the classification may include, but is not limited to: the classification may be performed by using an AHI of 15 as a classification limit, or may be performed by selecting another value, which is not limited in the present invention.

S203, using the image data of the detected object at five angles as the input of five image data obstructive sleep apnea detection models to obtain the corresponding outputs of the five models; meanwhile, the physiological data of the detected object is used as the input of the physiological data obstructive sleep apnea detection model, so that the corresponding output is obtained.

In specific implementation, dividing the processed samples into a training set, a verification set and a test set, carrying out neural network training, and obtaining a preliminary face detection OSA model through a training classifier; dividing the processed samples into a training set, a verification set and a test set, performing machine learning training, and obtaining a preliminary physiological data detection OSA model through a training classifier; and after the preliminary detection OSA model is obtained, the processed data is used as the input information of the human face model OSA, the output of the human face recognition OSA model processing image sample is obtained and used as the high-level representation sample of the corresponding image, and then the high-level representation sample of each sample is obtained. The middle layer output of the image data sample processed by the face detection OSA model is output through the characteristic layer of the primary face detection OSA model, and the characteristic of the face detection OSA model is output as the high-level representation sample of the corresponding sample.

And S204, performing weighted summation on the outputs obtained by the six models to obtain a final output, and obtaining a final prediction result. In the embodiment of the invention, the adopted human face detection obstructive sleep apnea disease model is obtained by training through a deep learning method by taking an apnea low ventilation index (AHI) value obtained by a corresponding patient wearing PSG equipment to monitor all night according to the existing picture and physiological data.

In specific implementation, the feature layer output of each face detection OSA model is used as a high-level representation subset to perform classifier training, and ensemble learning is performed to obtain a final prediction result.

Fig. 3 is a flowchart of model training provided in an embodiment of the present invention, and as shown in fig. 3, the flowchart specifically includes:

s301, image information is acquired.

In specific implementation, the head region of the detected object can be shot at, but not limited to, the following five preset angles to obtain a corresponding head image: information of front, side to left, side to right, 45 degrees to left, and 45 degrees to right angles.

S302, preprocessing the image.

In specific implementation, the head images of the head region of the detected object at five angles can be put into a face recognition model to obtain the face image of the detected object, and the whole head region of the detected object, including hair, ears and neck, is obtained by increasing a threshold value. Then, the picture is compressed to the longer side as 384 pixels according to the original length-width ratio in equal proportion, and then the fixed value is filled in the short side to make the length of the short side the same as that of the long side.

And S303, dividing a training set, a test set and a verification set, and performing image enhancement on the training set.

In specific implementation, the obtained physiological data are classified according to normal, mild, moderate and severe degrees, and a training set, a verification set and a test set are classified according to a preset proportion in each subclass. Optionally, in order to improve the generalization capability of the model, the sample image data of the model training may be further subjected to the following image enhancement processing to prevent overfitting: generating a countermeasure network, adjusting data training weight, increasing and decreasing sampling, and the like.

S304, training the model and obtaining the optimal solution.

In specific implementation, the face recognition model can be obtained by training the following neural network: a residual neural network, a convolutional neural network based on an inverted residual module, etc. And training the neural network by adjusting proper network hyper-parameters to obtain a face recognition model.

S305, outputs the probability corresponding to each category.

In specific implementation, the probability corresponding to each category is output, the output result of each model is the probability of each category, namely the probability of disease or no disease, but in practical application, the output of the neural network is a high-level feature sample corresponding to a picture.

Fig. 4 is a flowchart of training an image data obstructive sleep apnea detection model provided in an embodiment of the present invention, and as shown in the figure, the image data obstructive sleep apnea detection model specifically includes:

s401, acquiring a high-level sample.

In specific implementation, the picture can be put into a neural network, and the feature extraction layer of the picture is output to obtain a corresponding high-level sample.

S402, performing feature engineering on the high-level sample data.

In particular implementations, characterizing the physiological data includes, but is not limited to: the method comprises the steps of feature normalization, feature fusion, principal component analysis, local preserving projection, Laplace feature mapping, local linear embedding, linear discriminant analysis and the like.

And S403, dividing a training set, a test set and a verification set.

In specific implementation, the obtained physiological data are classified according to normal, mild, moderate and severe degrees, and a training set, a verification set and a test set are classified according to a preset proportion in each subclass.

S404, training the model and obtaining the optimal solution.

In particular implementations, training the model includes, but is not limited to, classification using machine learning algorithms, where the classifiers used may include, but are not limited to: xgboost, lightgmm, neural network, support vector machine, decision tree classifier, Bayes decision classifier.

S405, outputting the probability corresponding to each category.

In specific implementation, the corresponding probability of each category is output, and the output result of each model is the probability of each category, namely the probability of disease existence/disease nonexistence.

Fig. 5 is a flowchart of a training process of a physiological data obstructive sleep apnea detection model provided in an embodiment of the present invention, as shown in fig. 5, specifically including:

s501, acquiring physiological data.

In particular implementations, the acquired physiological data includes, but is not limited to: age, sex, neck circumference, BMI, etc. of the subject.

And S502, performing characteristic engineering on the physiological data.

In particular implementations, characterizing the physiological data includes, but is not limited to: the method comprises the steps of feature normalization, feature fusion, principal component analysis, local preserving projection, Laplace feature mapping, local linear embedding, linear discriminant analysis and the like.

S503, dividing a training set, a test set and a verification set.

In specific implementation, the obtained physiological data are classified according to normal, mild, moderate and severe degrees, and a training set, a verification set and a test set are classified according to a preset proportion in each subclass.

S504, training the model and obtaining the optimal solution.

In particular implementations, training the model includes, but is not limited to, classification using machine learning algorithms, where the classifiers used may include, but are not limited to: xgboost, lightgmm, neural network, support vector machine, decision tree classifier, Bayes decision classifier.

And S505, outputting the probability corresponding to each category.

In specific implementation, the corresponding probability of each category is output, and the output result of each model is the probability of each category, namely the probability of disease existence/disease nonexistence.

Based on the same inventive concept, the embodiment of the present invention further provides an obstructive sleep apnea detecting apparatus based on deep learning, as described in the following embodiments. Because the principle of the device for solving the problems is similar to the method for detecting the obstructive sleep apnea based on the deep learning, the implementation of the device can refer to the implementation of the method for detecting the obstructive sleep apnea based on the deep learning, and repeated parts are not repeated.

Fig. 6 is a schematic diagram of an obstructive sleep apnea detecting apparatus based on deep learning according to an embodiment of the present invention, as shown in fig. 6, the apparatus includes: a physiological data processing module 61, an image data processing module 62 and an obstructive sleep apnea detection module 63.

The physiological data processing module 61 is configured to input physiological data of a detected object into a pre-trained physiological data obstructive sleep apnea detection model, and output a corresponding apnea Hypopnea index AHI classification result; an image data processing module 62, configured to input image data of a detected object into a pre-trained image data obstructive sleep apnea detection model, and output a corresponding AHI classification result, where the image data includes: shooting a plurality of head images of the head area of a detected object from a plurality of preset angles, wherein the head images of each preset angle are input into different image data obstructive sleep apnea detection models; and the obstructive sleep apnea detecting module 63 is configured to input the AHI classification result output by the physiological data obstructive sleep apnea detecting model and the AHI classification result output by each image data obstructive sleep apnea detecting model into the integrated learning algorithm model, and output the AHI classification result of the detected object.

In one embodiment, as shown in fig. 6, the deep learning based obstructive sleep apnea detecting apparatus provided in the embodiment of the present invention may further include: a sample physiological data acquisition module 64 for acquiring sample image data, wherein the sample image data includes: shooting head areas of sample objects with different AHI labels from a plurality of preset angles to obtain a plurality of head images; the physiological data obstructive sleep apnea detection model training module 65 is used for dividing the sample physiological data into training physiological data, verification physiological data and test physiological data according to a preset proportion; according to the training physiological data, training any one of the following network models by using a deep learning method to obtain a physiological data obstructive sleep apnea detection model: the system comprises an Xgboost model, a Light GBM model, a neural network model, a support vector machine model, a decision tree classification model and a Bayesian decision classification model; verifying the physiological data obstructive sleep apnea detection model obtained by training according to the verified physiological data until the model accuracy rate meets the preset condition; and testing the physiological data obstructive sleep apnea detection model which meets the preset conditions according to the tested physiological data.

In one embodiment, as shown in fig. 6, the deep learning based obstructive sleep apnea detecting apparatus provided in the embodiment of the present invention may further include: a sample image data obtaining module 66 for obtaining sample image data, wherein the sample image data includes: shooting head areas of sample objects with different AHI labels from a plurality of preset angles to obtain a plurality of head images; the image data obstructive sleep apnea detection model training module 67 is used for dividing sample image data into training image data, verification image data and test image data according to a preset proportion; according to the training image data, training any one of the following network models by using a deep learning method to obtain a plurality of image data obstructive sleep apnea detection models corresponding to a plurality of preset angle head images: the system comprises an Xgboost model, a Light GBM model, a neural network model, a support vector machine model, a decision tree classification model and a Bayesian decision classification model; verifying each image data obstructive sleep apnea detection model obtained by training according to verification image data until the model accuracy rate meets a preset condition; and testing the image data obstructive sleep apnea detection model which meets the preset conditions according to the test image data.

In one embodiment, as shown in fig. 6, the deep learning based obstructive sleep apnea detecting apparatus provided in the embodiment of the present invention may further include: a face recognition module 68, configured to input image data of the detected object into a pre-trained face recognition model, and output corresponding face image data, where the face recognition model is a model obtained by training a neural network model; the image data processing module 62 is further configured to input the face image data output by the face recognition model into a pre-trained image data obstructive sleep apnea detection model, and output a corresponding AHI classification result.

Optionally, the device for detecting obstructive sleep apnea based on deep learning provided in the embodiment of the present invention further includes: a face recognition model training module 69 for obtaining sample image data, wherein the sample image data includes: shooting head areas of sample objects with different AHI labels from a plurality of preset angles to obtain a plurality of head images; performing cluster analysis on head images with different preset angles in sample image data to obtain a high-level representation subset belonging to each type of OSA; training the following neural network model according to the high-level representation subset belonging to each type of OSA to obtain a face recognition model belonging to each type of OSA: a residual error neural network model and a convolution neural network model based on an inverted residual error module.

In an embodiment, in the deep learning based obstructive sleep apnea detecting apparatus provided in the embodiment of the present invention, the AHI classification results output by the physiological data obstructive sleep apnea detecting model and the image data obstructive sleep apnea detecting models are probabilities belonging to different AHI classifications, and the integrated learning algorithm model is configured to perform weighted summation on the probabilities that the detected object output by the physiological data obstructive sleep apnea detecting model belongs to different AHI classifications and the probabilities that the detected object output by the image data obstructive sleep apnea detecting models belongs to different AHI classifications, so as to obtain the AHI classification result of the detected object.

Based on the same inventive concept, a computer device is further provided in the embodiments of the present invention to solve the technical problem of complicated operation of the existing method for diagnosing OSA based on the sleep records of a patient all night, fig. 7 is a schematic diagram of a computer device provided in the embodiments of the present invention, as shown in fig. 7, the computer device 70 includes a memory 701, a processor 702, and a computer program stored on the memory 701 and operable on the processor 702, and when the processor 702 executes the computer program, the method for detecting obstructive sleep apnea based on deep learning is implemented.

Based on the same inventive concept, the embodiment of the present invention further provides a computer-readable storage medium for solving the technical problem of tedious operation of the existing method for diagnosing OSA based on the sleep records of the patient all night, and the computer-readable storage medium stores a computer program for executing the above-mentioned method for detecting obstructive sleep apnea based on deep learning.

To sum up, the embodiments of the present invention provide a method, an apparatus, a computer device, and a computer readable storage medium for obstructive sleep apnea detection based on deep learning, after acquiring physiological data of a detected subject and a plurality of head images obtained by shooting a head region of the detected subject from a plurality of angles, inputting the physiological data of the detected subject into a pre-trained physiological data obstructive sleep apnea detection model, outputting a corresponding AHI classification result, inputting the plurality of head images of the detected subject into pre-trained image data obstructive sleep apnea detection models corresponding to different preset angles, outputting a corresponding AHI classification result, and finally outputting the AHI classification result output by the physiological data obstructive sleep apnea detection model and the AHI classification result output by each image data obstructive sleep apnea detection model, inputting the result into an ensemble learning algorithm model, and outputting an AHI classification result of the detected object.

Compared with the technical scheme for diagnosing the OSA based on the night sleep record of the patient in the prior art, the obstructive sleep apnea detecting scheme provided by the embodiment of the invention only needs to collect the head images of the patient at multiple angles, does not need the user to wear various PSG (particle swarm optimization) devices, is simple to operate, and can detect the obstructive sleep apnea condition with a low AHI (advanced high-performance instrumentation index) value, so that the user can treat the OSA in time at the early stage.

In the embodiment of the invention, based on a deep learning technology, the AHI of the detected object is predicted by combining the head images of the detected object at a plurality of angles and physiological data, so as to judge whether the detected object has OSA. Compared with the scheme of carrying out OSA prediction by utilizing PSG of the detected object all night in the prior art, the scheme provided by the embodiment of the invention can facilitate the user to know the condition of an illness in time, treat the illness early, is convenient and fast, is easy to operate because the detected person does not need to wear various PSG devices, and has the recognition accuracy rate reaching 83 percent after verification.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are only exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.