Obstructive sleep apnea crowd AI screening model training system and method
Through AI screening models based on demographic and anthropometric data, the complex and costly problem of obstructive sleep apnea screening in the prior art is solved, and simpler and more efficient large-scale population screening is achieved.
Patent Information
- Application Number
- CN202411858353.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-04-11
AI Technical Summary
In the prior art, screening methods for obstructive sleep apnea require complex equipment and high-cost polysomnography monitoring, and can only be performed for single-person testing, and large-scale population screening cannot be achieved.
An AI screening model training system based on demographic feature data and anthropometric feature data is adopted, and a sample collection component and an AI training server are used to train and save the AI feature data set for obstructive sleep message pause.
It provides a simpler, more efficient, and suitable for large-scale population screening, which can accurately screen through basic data, reducing detection costs and complexity.
Smart Images
Figure CN119811662A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the technical field of screening and analysis of obstructive sleep apnea populations, and specifically relates to an AI screening model training system and method for obstructive sleep apnea populations. Background Art
[0002] Obstructive sleep apnea generally refers to an adult who has more than 30 episodes of sleep during a 7-hour sleep period each night. During each episode, the airflow from the mouth and nose stops for 10 seconds or longer, accompanied by a decrease in blood oxygen saturation. When breathing normally, the key to gas exchange is the upper respiratory tract above the larynx, which allows the gas to enter the trachea and bronchi smoothly. If for some reason, this airflow is blocked, snoring or obstructive sleep apnea will occur. From an anatomical point of view, there are three parts above the larynx that are prone to stenosis and obstruction: the nose and nasopharynx, the oropharynx and soft palate, and the root of the tongue.
[0003] Sleep Apnea Hypopnea Syndrome (SAHS) refers to a clinical syndrome in which various reasons lead to repeated apnea and / or hypopnea, hypercapnia, and sleep interruption during sleep, resulting in a series of pathophysiological changes in the body.
[0004] Sleep apnea-hypopnea syndrome refers to repeated episodes of apnea more than 30 times per night during sleep or an apnea-hypopnea index (AHI) ≥ 5 times / hour, accompanied by clinical symptoms such as drowsiness. Apnea refers to the complete cessation of oral and nasal airflow for more than 10 seconds during sleep; hypopnea refers to a decrease in the intensity (amplitude) of respiratory airflow by more than 50% compared to the baseline level during sleep, accompanied by a decrease in blood oxygen saturation by ≥ 4% compared to the baseline level or micro-arousals; the apnea-hypopnea index refers to the number of apneas plus hypopneas per hour of sleep.
[0005] Sleep apnea hypopnea syndrome includes: 1. Central type (CSAS) 2. Obstructive type (OSAS) 3. Mixed type (MSAS). In recent years, with the continuous progress of various testing methods, in-depth and comprehensive research on obstructive type (OSAS) has found that this disease is complex, changeable and ubiquitous. Therefore, obstructive type (OSAS) cannot be generally regarded as an annoying sound that affects the rest of others in social life, but a clinical disease that needs to be carefully examined. It can cause many serious complications. Since the anatomical part of snoring is closely related to the ear, nose and throat, many patients often seek treatment from the otolaryngology department first.
[0006] In the prior art, polysomnography (PSG) is the most important examination for diagnosing sleep snoring (sleep apnea-hypopnea syndrome). Through the continuous monitoring of breathing, arterial oxygen saturation, electroencephalogram, electrocardiogram, heart rate and other indicators at night, it is possible to understand whether the snorer has respiratory arrest, the number of pauses, the duration of the pauses, the lowest arterial oxygen value when the pauses occur, and the degree of impact on physical health. It is the internationally recognized gold standard for diagnosing sleep apnea-hypopnea syndrome. Polysomnography is the most commonly used sleep monitoring method, the most important examination for diagnosing snoring, and the internationally recognized gold standard for diagnosing sleep apnea-hypopnea syndrome.
[0007] like Fig.11 As shown, wearing various sensors on the body during sleep at night will have an adverse effect on sleep itself, and the cost of detection is also very high, requiring comprehensive analysis of various equipment and various data. Moreover, it can only be tested on a single person, and cannot be promoted on a large scale, let alone achieve large-scale screening of the population. How to conduct obstructive sleep apnea screening for the population in a more convenient and simple way is a problem to be solved.
[0008] Technical Glossary:
[0009] LGB model: LightGBM (lgb model) is a machine learning algorithm based on the gradient boosting framework, mainly used for the optimization and training of decision trees.
[0010] LGB: Lightweight Gradient Boosting Machine.
[0011] XGBoost: Extreme Gradient Boosting.
[0012] RF: Random Forest.
[0013] FNN: Feedforward neural network.
[0014] CNN: Convolutional Neural Network.
[0015] LOR: Logistic regression.
[0016] LOG2: Logistic regression model 2 established after dealing with multicollinearity variables. Summary of the invention
[0017] In the present application, the inventors propose a system and method for training an AI screening model for people with obstructive sleep apnea based on demographic characteristic data and anthropometric characteristic data, which avoids the inconvenience of requiring complex equipment to perform measurements during sleep during the initial screening process, and provides a more convenient, efficient and simple AI screening system and method for people with obstructive sleep apnea that is suitable for population screening.
[0018] A training system for an AI screening model for an obstructive sleep apnea population, comprising a sample collection component and an AI training server; the sample collection component comprises a sample annotation module; a sample input component collects sample data, wherein the sample data is physical characteristic parameters of patients diagnosed with obstructive sleep apnea and / or patients diagnosed with non-obstructive sleep apnea; the sample annotation module converts the physical parameter data into standard AI training input data; the AI training input data is input into the AI training server, and an AI characteristic data set for obstructive sleep apnea is obtained after training with the AI training server; the physical characteristic parameters comprise general demographic characteristic data and anthropometric characteristic data, and the anthropometric characteristic data comprise height, weight, waist circumference, neck circumference, and hip circumference.
[0019] The general demographic characteristic data may include age and gender.
[0020] It can be that the sample annotation module converts the body parameter data into standard AI training input data.
[0021] It may be that the physical characteristic parameter includes triglyceride content.
[0022] It may be that the physical characteristic parameter includes high-density lipoprotein cholesterol level.
[0023] The AI training input data may include any one or more of body mass index (BMI), lipid accumulation index (LAP), visceral fat (VAI), Chinese visceral fat index (CVAI), body shape index (ABSI), body roundness index (BRI), body fat estimate (BAE), body build index (PI), relative fat mass (RFM), taper index (CI), abdominal volume index (AVI), waist-to-hip ratio (WHR), waist-to-height ratio (WHtR), anthropometric prediction equation, predicted lean body mass (kg), predicted body fat mass (Fat mass (kg), predicted body fat percentage (%), and hip circumference (HC).
[0024] It may be that the AI training server includes one or more algorithm modules, and the algorithm modules include a machine learning algorithm module, a deep learning module, and a logistic regression calculation module; the AI training server saves the obstructive sleep apnea AI feature data set corresponding to the algorithm module into a database according to different algorithm modules.
[0025] It may be that the AI training server is connected to the sample collection component via a network.
[0026] A method for training an AI screening model for a population with obstructive sleep apnea, comprising collecting sample data, wherein the sample data is physical characteristic parameters of patients diagnosed with obstructive sleep apnea and / or patients diagnosed with non-obstructive sleep apnea; converting the physical parameter data into standard AI training input data; inputting the AI training input data into an AI training server, and obtaining an obstructive sleep apnea AI characteristic data set after training with the AI training server; the physical characteristic parameters include general demographic characteristic data and anthropometric characteristic data, and the anthropometric characteristic data include height, weight, waist circumference, neck circumference, and hip circumference.
[0027] It can be that the general demographic characteristic data includes age and gender; TB2: the sample annotation module converts the body parameter data into standard AI training input data; TB3: the body characteristic parameters include triglyceride content; TB4: the body characteristic parameters include high-density lipoprotein content.
[0028] The AI training input data may include any one or more of body mass index (BMI), lipid accumulation index (LAP), visceral fat (VAI), Chinese visceral fat index (CVAI), body shape index (ABSI), body roundness index (BRI), body fat estimate (BAE), body build index (PI), relative fat mass (RFM), taper index (CI), abdominal volume index (AVI), waist-to-hip ratio (WHR), waist-to-height ratio (WHtR), anthropometric prediction equation, predicted lean body mass (kg), predicted body fat mass (Fat mass (kg), predicted body fat percentage (%), and hip circumference (HC).
[0029] It may be that the AI training server includes one or more algorithm modules, and the algorithm modules include a machine learning algorithm module, a deep learning module, and a logistic regression calculation module; the AI training server saves the obstructive sleep apnea AI feature data set corresponding to the algorithm module into a database according to different algorithm modules.
[0030] It may be that, during the process of collecting sample data, anthropometric characteristic data measurement is performed on sample objects that lack physical characteristic parameters.
[0031] One of the technical effects of the above technical solution is: the AI screening model training system and method for obstructive sleep apnea population is based on demographic characteristic data and anthropometric characteristic data, providing a simpler and more efficient screening tool and method.
[0032] One of the technical effects of the above technical solution is that basic data such as age, gender, height, weight, waist circumference, neck circumference and hip circumference are relatively easy to obtain, and after a large amount of experimental data analysis and verification, they meet the measurement accuracy required for detection or screening.
[0033] One of the technical effects of the above technical solution is that the sample annotation module converts the body parameter data into standard AI training input data, and the standard AI training input data obtained during the conversion accumulates data for subsequent research.
[0034] One of the technical effects of the above technical solution is that the body characteristic parameters including polysomnography PSG data can provide a more accurate reference and improve the accuracy of training.
[0035] One of the technical effects of the above technical solution is that the algorithm module includes a variety of algorithms with good adaptability, and proves that data based on age, gender, height, weight, waist circumference, neck circumference, and hip circumference are strongly correlated with obstructive sleep apnea, and have little to do with the choice of AI model. The most relevant parameters can be selected through the above data, and it has good adaptability to different types of people.
[0036] One of the technical effects of the above technical solution is that the AI training server is connected to the sample collection component through the network, which can improve the data collection ability of the AI training server. The more data there is, the better the adaptability of the trained model and data set. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is a schematic block diagram of an AI screening model training system for obstructive sleep apnea population;
[0038] Figure 2 It is a schematic block diagram of an AI screening model training system for obstructive sleep apnea population;
[0039] Figure 3 It is a schematic block diagram of an AI screening model training system for obstructive sleep apnea population;
[0040] Figure 4 It is a schematic block diagram of an AI screening model training system for obstructive sleep apnea population;
[0041] Figure 5 It is a schematic block diagram of an AI screening model training system for obstructive sleep apnea population;
[0042] Figure 6 It is a schematic block diagram of an AI screening model training system for obstructive sleep apnea population;
[0043] Figure 7 It is a schematic diagram of the AI screening steps for people with obstructive sleep apnea;
[0044] Figure 8 This is a schematic diagram of the interface for training an AI screening model for people with obstructive sleep apnea;
[0045] Fig. 9 It is the training result of clinical dataset;
[0046] Fig.10 It is the training result of clinical dataset;
[0047] Fig.11 It is a schematic diagram of obstructive sleep apnea detection in the prior art. DETAILED DESCRIPTION
[0048] The content of this application is further described in detail below in conjunction with the accompanying drawings. It should be noted that the following is a description of the preferred embodiments of the present invention and does not constitute any limitation to the present invention. The description of the preferred embodiments of the present invention is only an illustration of the general principles of the present invention. The numbers such as "first", "second" and "A" and "B" involved in the present invention are only for the convenience of explanation and do not represent the order relationship in time or space. The combination of letters and numbers "TA", "TB" and "H" involved in the present invention are only for the convenience of explanation, and the specific meaning is determined by the specific words referred to.
[0049] like Figure 1 and Figure 2 , an AI screening model training system for obstructive sleep apnea population, comprising a sample collection component and an AI training server; the sample collection component comprises a sample annotation module; the sample input component collects sample data, the sample data being physical characteristic parameters of patients diagnosed with obstructive sleep apnea and / or patients diagnosed with non-obstructive sleep apnea; the sample annotation module converts the physical parameter data into standard AI training input data; the AI training input data is input into the AI training server, and an obstructive sleep apnea AI feature data set is obtained after training by the AI training server; the physical characteristic parameters include general demographic characteristic data and anthropometric characteristic data, and the anthropometric characteristic data include height, weight, waist circumference, neck circumference, and hip circumference.
[0050] like Figure 2 , an AI screening model training system for obstructive sleep apnea, wherein the general demographic characteristic data includes age and gender. The physical characteristic parameters include polysomnography PSG data. Figure 2 The AI training server is connected to the sample collection component through a network. The demographic data, anthropometric data and PSG data are transmitted to the terminal data collector through the Internet of Things, and then transmitted to the AI training server through the local area network. The training is performed on the server and the data is stored in the database.
[0051] like Figure 3 , an AI screening model training system for people with obstructive sleep apnea. The original data acquisition device is transmitted to the cloud platform via wireless communication for data conversion and analysis and processing and archived in the database.
[0052] The AI training server includes one or more algorithm modules, including a machine learning algorithm module, a deep learning module, and a logistic regression calculation module; the AI training server saves the obstructive sleep apnea AI feature data set corresponding to the algorithm module into a database according to different algorithm modules.
[0053] like Figure 3 , an AI screening model training system for obstructive sleep apnea. In the AI training server, data is preprocessed after reading, and then feature engineering is performed to convert it into standard AI training input data. The standard AI training input data is input into model training, and model selection and parameter tuning are performed after model evaluation to finally generate a model. Through model selection and parameter tuning, the most suitable algorithm module for this application can be selected.
[0054] like Figure 5 to Figure 6 In an embodiment of the training method of the AI screening model for obstructive sleep apnea population, sample data is collected, and the sample data is the physical characteristic parameters of patients diagnosed with obstructive sleep apnea and / or patients diagnosed with non-obstructive sleep apnea; the physical parameter data is converted into standard AI training input data; the AI training input data is input into the AI training server, and the obstructive sleep apnea AI feature data set is obtained after training with the AI training server; the physical characteristic parameters include general demographic characteristic data and anthropometric characteristic data, and the anthropometric characteristic data include height, weight, waist circumference, neck circumference, and hip circumference. The general demographic characteristic data include age and gender. The physical characteristic parameters include polysomnography PSG data. The AI training server includes one or more algorithm modules, and the algorithm modules include machine learning algorithm modules, deep learning modules, and logistic regression calculation modules; the AI training server uses different algorithm modules to calculate the physical characteristic parameters.
[0055] The obstructive sleep apnea AI feature data set corresponding to the algorithm module is saved in a database.
[0056] like Figure 4In the figure, (A) AUC plot of the training set of clinical experimental dataset 1; (B) AUC plot of the validation set of clinical experimental dataset 1; (C) AUC plot of clinical experimental dataset 2; (D) AUC plot of the community-based sleep heart health research dataset; (E) Calibration plot of the training set of clinical experimental dataset 1; (F) Calibration plot of the validation set of clinical experimental dataset 1; (G) Calibration plot of clinical experimental dataset 2; (H) Calibration plot of the community-based sleep heart health research dataset; (I) Net benefit analysis decision curve of clinical experimental dataset 1; (J) Net benefit analysis decision curve of clinical experimental dataset 1 validation set; (K) Net benefit analysis decision curve of clinical experimental dataset 2; (L) Net benefit analysis decision curve of the community-based sleep heart health research dataset. LGB: lightweight gradient boosting machine; XGBoost: extreme gradient boosting; RF: random forest; FNN: feedforward neural network; CNN: convolutional neural network; LOR: logistic regression. LOG2: logistic regression model 2 established after dealing with multicollinear variables.
[0057] like Figure 5 , collect sample data, including the following steps: initial screening of test persons, questionnaire survey of test persons, measurement of test persons' physical parameters, polysomnography, and distinguishing obstructive sleep apnea patients and / or non-obstructive sleep apnea patients through the above process; AI model training is performed with the body feature parameters with the above marks to obtain an obstructive sleep apnea AI feature data set.
[0058] like Figure 6 , collecting sample data, including the following steps: initial screening of cases, finding confirmed obstructive sleep apnea patients and / or non-obstructive sleep apnea patients, and performing supplementary measurements of body parameters to obtain body parameters. That is, in the process of collecting sample data, anthropometric characteristic data are measured for sample subjects that lack physical characteristic parameters.
[0059] like Figure 7 , is in practice, on the clinical side, 11,224 participants were collected for initial data collection, of which 2,893 participants’ data were excluded, and the final training analysis used 8,331 participants’ data. On the community side, multiple data were collected, of which 1,119 participants’ data were excluded, and 4,685 participants’ data were used as test sample data. The above data were input into a machine learning-based model, namely the moderate to severe OSA screening model, for training to obtain the training experimental data set 1. In subsequent use and testing, clinical data sets and community data sets were continuously collected, and training and screening were carried out at the same time.
[0060] like Figure 8, is one of the running interfaces of the above software. It can be seen that the input data includes age, gender, height, weight, waist circumference, hip circumference, and neck circumference.
[0061] like Figure 8 , is one of the software operation interfaces mentioned above. It can be seen that the input data also includes the content values of triglycerides (mmol / L) and high-density lipoprotein cholesterol (mmol / L).
[0062] like Figure 8 , which is one of the software operation interfaces mentioned above. It can be seen that the input data of standard AI training also includes a variety of obesity indicators.
[0063] The body parameter data is converted into standard AI training input data, where the AI training input data includes:
[0064] 1. Body mass index (BMI) is the body mass index (body
[0065] mass index, BMI);
[0066] 2.Lipid accumulation product (LAP);
[0067] 3. Visceral adiposity index (VAI);
[0068] 4. Chinese visceral adiposity index (Chinese visceral adiposity index,
[0069] CVAI);
[0070] 5. Body Shape Index (ABSI);
[0071] 6. Body roundness index (BRI);
[0072] 7. Body fat estimator, also known as body adiposity estimator (BAE);
[0073] 8. Ponderal Index (PI);
[0074] 9.Relative fat mass (RFM);
[0075] 10. Conicity index is the cone cutting index (CI);
[0076] 11. Abdominal volume index (AVI);
[0077] 12. Waist-to-Hip Ratio (WHR);
[0078] 13. Waist-to-Height Ratio (WHtR);
[0079] 14. Anthropometric prediction equation;
[0080] 15. Predicted lean body mass (kg);
[0081] 16. Predict body fat mass (kg);
[0082] 17. Predicted body fat percentage Percent fat (%);
[0083] 18. Hip circumstance (HC);
[0084] 19. Body adiposity index (BAI).
[0085] Waist-to-Height Ratio (WHtR) refers to the waist circumference divided by the height.
[0086] Waist-to-Hip Ratio (WHR) = Waist circumference / Hip circumference. Waist circumference or minimum waist circumference is used when measuring, and waist circumference (umbilicus) is not used.
[0087] Waist-to-hip ratio (WHR),
[0088] Body mass index (BMI): also known as body mass index or body mass index, this index is calculated by dividing weight (kg) by height (m) squared. The ratio obtained by this formula can reflect the body density to a certain extent. The obesity index equation is the calculation formula of BMI (Body Mass Index).
[0089] Ponderal Index: The ratio of weight to height. This index is the body mass index (ponderal index = W / h 3 ); body mass index (ponderal index, PI),
[0090] Conicity index (CI),
[0091] Relative fat mass index (RFM). You only need to know your height and waist circumference to estimate your body fat content. Relative fat mass index (RFM), RFM = 64-(20×(Height / Waist))+(12×Sex) (male = 0; female = 1).
[0092] Visceral fat (VAI) is a type of body fat. It is different from subcutaneous fat (the "fat" that we can feel on the body). It surrounds the internal organs and is mainly found in the abdominal cavity. Visceral fat is of great significance to our health.
[0093] Abdominal volume index (AVI); calculation formula
[0094]
[0095] Lipid accumulation product (LAP), in 2005, Kahn et al. proposed the concept of lipid accumulation product (LAP), which combines the anatomical index WC reflecting excessive body fat accumulation and the physiological index triglyceride (TG), and then transforms it into a continuous variable according to gender. It was found that LAP is significantly better than body mass index (BMI) in identifying the risk of diabetes and cardiovascular disease, and can better reflect visceral lipid accumulation. 男 =[WC-65]×TG(cm·mmol / L); LAP 女 =[WC-58]×TG (cm·mmol / L).
[0096] Visceral adiposity index (VAI),
[0097] male: female:
[0098] The Chinese visceral adiposity index (CVAI) is an index established based on VAI (visceral fat index) to better reflect the distribution and metabolism of visceral fat in Chinese people. Visceral fat index (China) (Chinese visceral adiposity index, CVAI);
[0099] male:
[0100] CVAI=-267.93+0.68*Age+0.03*BMI+4.00*WC+22.00*Log 10 TG-16.32*HDL-C;
[0101] female:
[0102] CVAI=-187.32+1.71*Age+4.23*BMI+1.12*WC+39.76*Log 10 TG-11.66*HDL-C.
[0103] In the above formula, "*" means "×".
[0104] A Body Shape Index (ABSI) is a new body shape evaluation indicator. In addition to taking into account the traditional body mass index (BMI) data, it also needs to consider waist circumference. The use of this ABSI indicator to evaluate and predict the risk of premature death is better than the traditional BMI indicator.
[0105] Body roundness index (BRI). This index is estimated based on height and WC and can assess the content of visceral fat. The higher the value, the more obvious the accumulation of visceral fat and the higher the incidence of obesity-related diseases. The “*” in the formula means “×”.
[0106] body adiposity estimator (BAE),
[0107] (Male = 0; Female = 1).
[0108] Predict lean body mass (LBM),
[0109] male:
[0110] LBM=19.363+0.001×Age+0.064×Height
[0111] +0.756×Weight-0.366×Waist-0.066×Mexican
[0112] +0.231×Hispanic+0.432×Blanck-1.007×Other Ethnicity(kg) Female:LBM=-10.683-0.039×Age+0.186×Height
[0113] +0.383×Weight-0.043×Waist-0.359×Mexican-0.059×Hispanic
[0114] +1.085×Blanck-0.34×Other Ethnicity(kg).
[0115] Predicted body fat (fat mass, FM) (kg),
[0116] Male: FM=-18.592-0.009×Age-0.080×Height
[0117] +0.226×Weight+0.387×Waist+0.080×Mexican
[0118] -0.188×Hispanic-0.483×Blanck+1.050×Other Ethnicity(kg)Female: FM=11.817+0.041×Age-0.199×Height
[0119] +0.610×Weight+0.044×Waist+0.388×Mexican
[0120] +0.073×Hispanic-1.187×Blanck+0.325×Other Ethnicity(kg).
[0121] Predicted body fat percentage (percent fat, PF),
[0122] Male: PF=0.02+0.00×Age-0.07×Height-0.08×Weight
[0123] +0.48×Waist+0.32×Mexican+0.02×Hispanic-0.65×Blanck
[0124] +1.12×Other Ethnicity(%);
[0125] Female: PF=50.46+0.07×Age-0.26×Height+0.27×Weight
[0126] +0.10×Waist+0.89×Mexican+0.49×Hispanic-1.57×Blanck
[0127] +0.43×Other Ethnicity(%).
[0128] like Fig. 9 As shown, sub-figure A is the AUC graph of the training set based on the clinical experimental dataset 1; sub-figure B is the AUC graph of the validation set based on the clinical experimental dataset 1; sub-figure C is the AUC graph of the clinical based on the experimental dataset 2; sub-figure D is the AUC graph of the community based on the sleep heart health research dataset; sub-figure E is the calibration graph of the training set based on the clinical experimental dataset 1; sub-figure F is the calibration graph of the validation set based on the clinical experimental dataset 1; sub-figure G is the calibration graph of the clinical based on the experimental dataset 2; sub-figure H is the calibration graph of the community based on the sleep heart health research dataset; sub-figure I is the net benefit analysis decision curve of the clinical based on the experimental dataset 1; sub-figure J is the net benefit analysis decision curve of the validation set based on the clinical experimental dataset 1; sub-figure K is the net benefit analysis decision curve of the clinical based on the experimental dataset 2; sub-figure L is the net benefit analysis decision curve of the community based on the sleep heart health research dataset.
[0129] like Fig.10 As shown, sub-figure A is the training set AUC curve of clinical experiment data set 1; sub-figure B is the validation set AUC curve of clinical experiment data set 1; sub-figure C is the training set calibration curve of clinical experiment data set 1; sub-figure D is the validation set calibration curve of clinical experiment data set 1; sub-figure E is the training set net benefit analysis decision curve of clinical experiment data set 1; sub-figure F is the validation set net benefit analysis decision curve of clinical experiment data set 1.
[0130] Fig.10In the LGB model 1, the variables used as input included NC, HC, WC, BMI, ABSI, BRI, BAE, PI, RFM, CI, AVI, WHR, WHtR, predicted lean body mass, predicted fat mass, predicted fat percentage, LAP, VAI, and CVAI.
[0131] Fig.10 In the LGB model 2, the variables used as input included NC, HC, WC, BMI, ABSI, BRI, BAE, PI, RFM, CI, AVI, WHR, WHtR, predicted lean body mass, predicted fat mass, predicted fat percentage, LAP, VAI, CVAI, TG, and HDL.
[0132] Fig.10 In the LGB model 3, the variables used as inputs included NC, HC, WC, BMI, ABSI, BRI, BAE, PI, RFM, CI, AVI, WHR, WHtR, predicted lean body mass, predicted fat mass, and predicted fat percentage.
[0133] Fig.10 In the LGB model 4, the variables used as inputs included NC, HC, WC, BMI, ABSI, BRI, BAE, PI, RFM, CI, AVI, WHR, WHtR, predicted lean body mass, predicted fat mass, predicted fat percentage, TG, and HDL.
[0134] Although the present invention is illustrated and described according to the preferred embodiment and several alternatives, the invention is not limited by the specific description in this specification. Other additional replacement or equivalent components can also be used to practice the present invention.
Claims
1. An AI screening model training system for obstructive sleep apnea, characterized in that : Includes sample collection components and AI training servers; The sample collection component includes a sample annotation module; The sample input component collects sample data, wherein the sample data is a physical characteristic parameter of a patient diagnosed with obstructive sleep apnea and / or a patient diagnosed with non-obstructive sleep apnea; The sample annotation module converts the body parameter data into standard AI training input data; the AI training input data is input into the AI training server, and the obstructive sleep apnea AI feature data set is obtained through training by the AI training server; The physical characteristic parameters include general demographic characteristic data and anthropometric characteristic data. The anthropometric characteristic data include height, weight, waist circumference, neck circumference, and hip circumference.
2. The obstructive sleep apnea population AI screening model training system according to claim 1, characterized in that: Include any of the following technical features: TA1: The general demographic characteristics data include age and gender; TA2: The sample annotation module converts the body parameter data into standard AI training input data; TA3: The physical characteristic parameters include triglyceride content; TA4: The physical characteristic parameters include high-density lipoprotein cholesterol content.
3. The obstructive sleep apnea population AI screening model training system according to claim 1, characterized in that: The AI training input data Including body mass index (BMI); lipid accumulation index (LAP); visceral fat (VAI); Chinese visceral fat index (CVAI); body shape index (ABSI); body roundness index (BRI); body fat estimation (BAE); and body mass index (PI). Relative fat mass RFM; taper index CI; abdominal volume index AVI; waist-to-hip ratio WHR; waist-to-height ratio WHtR; anthropometric prediction equation; predicted lean body mass (kg); predicted fat mass (kg); predicted percent fat (%); any one or more of the following: hip circumference HC.
4. The obstructive sleep apnea population AI screening model training system according to claim 1, characterized in that: The AI training server includes one or more algorithm modules, including a machine learning algorithm module, a deep learning module, and a logistic regression calculation module; the AI training server saves the obstructive sleep apnea AI feature data set corresponding to the algorithm module into a database according to different algorithm modules.
5. The obstructive sleep apnea population AI screening model training system according to claim 1, characterized in that: The AI training server is connected to the sample collection component via a network.
6. A method for training an AI screening model for obstructive sleep apnea, characterized in that: Collecting sample data, wherein the sample data is physical characteristic parameters of patients diagnosed with obstructive sleep apnea and / or patients diagnosed with non-obstructive sleep apnea; Converting the body parameter data into standard AI training input data; Inputting the AI training input data into an AI training server, and obtaining an obstructive sleep apnea AI feature data set through training on the AI training server; The physical characteristic parameters include general demographic characteristic data and anthropometric characteristic data. The anthropometric characteristic data include height, weight, waist circumference, neck circumference, and hip circumference.
7. The obstructive sleep apnea population AI screening model training method according to claim 1, characterized in that: Include any of the following technical features: TB1: The general demographic characteristics data include age and gender; TB2: The sample annotation module converts the body parameter data into standard AI training input data; TB3: The body characteristic parameters include triglyceride content; TB4: The body characteristic parameters include high-density lipoprotein cholesterol content.
8. The obstructive sleep apnea population AI screening model training method according to claim 1, characterized in that: The AI training input data, Including body mass index (BMI); lipid accumulation index (LAP); visceral fat (VAI); Chinese visceral fat index (CVAI); body shape index (ABSI); body roundness index (BRI); body fat estimation (BAE); and body mass index (PI). Relative fat mass RFM; taper index CI; abdominal volume index AVI; waist-to-hip ratio WHR; waist-to-height ratio WHtR; anthropometric prediction equation; predicted lean body mass (kg); predicted fat mass (kg); predicted percent fat (%); any one or more of the following: hip circumference HC.
9. The obstructive sleep apnea population AI screening model training method according to claim 6, characterized in that: The AI training server includes one or more algorithm modules, including a machine learning algorithm module, a deep learning module, and a logistic regression calculation module; the AI training server saves the obstructive sleep apnea AI feature data set corresponding to the algorithm module into a database according to different algorithm modules.
10. The obstructive sleep apnea population AI screening model training method according to claim 6, characterized in that: During the sample data collection process, anthropometric characteristic data are measured for sample objects that lack physical characteristic parameters.