A hierarchical diabetes risk assessment method based on big data and deep learning
This hierarchical diabetes risk assessment method, which combines big data and deep learning with a one-dimensional convolutional neural network and attention mechanism, solves the problems of insufficient accuracy and personalization in existing models. It achieves high-precision diabetes risk assessment and early warning, and is suitable for various application scenarios and individual health management.
Patent Information
- Application Number
- CN202510478466.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-16
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2045-04-16
AI Technical Summary
Existing diabetes risk assessment models are inadequate in terms of accuracy, application scenarios, personalization, and data integration, and cannot meet the needs of large-scale application and personalized health management.
A hierarchical diabetes risk assessment method based on big data and deep learning is adopted. It combines a one-dimensional convolutional neural network and attention mechanism, classifies risk levels through a normal distribution mapping model, and uses a data perturbation method to rank the importance of single-sample features to achieve personalized diabetes risk assessment.
It provides high-precision, personalized diabetes risk assessment, enabling early warning and accurate health management, and is suitable for different application scenarios and individual health data.
Smart Images

Figure CN120376146B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of big data, specifically relating to a hierarchical diabetes risk assessment method based on big data and deep learning. Background Technology
[0002] Early diagnosis, early treatment, and timely lifestyle modifications are the most effective ways to prevent and control diabetes and its complications. However, due to limitations in economic status and understanding of their own health, only half of people with diabetes are currently diagnosed with the disease, posing a challenge to early detection and treatment.
[0003] How can accurate and effective early diagnosis of diabetes be achieved? Domestic health and medical big data has become a crucial entry point. Health and medical big data is a vital strategic resource for my country, specifically including health big data centered on physical examination data, medical big data centered on electronic medical records, examinations, tests, and prescriptions, rapidly growing biomics big data, and public health big data based on regional population disease and health status monitoring. In recent years, physical examinations have received increasing attention, and physical examination institutions and data platforms have developed rapidly, with the function of these institutions gradually shifting from simple health checkups to health management. For the potential diabetes population in China, physical examinations offer advantages such as convenience, wide coverage, and comprehensive examination items. Therefore, physical examination big data is showing an exponential growth trend. However, physical examination indicators, as an important tool for early detection of disease risk factors, have not yet been fully utilized to establish domestic chronic disease risk assessment tools. With the rapid development of artificial intelligence, deep learning algorithms can leverage their performance advantages on big data, and machine learning can help understand and quantify the relationship between various features and diseases. By combining artificial intelligence to fully utilize risk factors in physical examination data to build disease risk assessment tools, we can not only provide accurate, convenient and universal risk warnings for potential patients, but also directly apply them to people undergoing physical examinations and health management centers. In addition, they can also serve as a tool for medical insurance systems and other entities to assess individual disease risks. This has great practical significance and broad application prospects, and is of great importance for carrying out personalized health management and early disease prevention.
[0004] Currently, numerous studies both domestically and internationally have developed diabetes risk models. However, different countries have varying lifestyles and environments, requiring each country to develop its own diabetes risk scoring model tailored to its specific population. China has also developed some diabetes risk assessment tools based on its own population sample and physical characteristics. However, the application scenarios for existing risk prediction or risk scoring models are not clearly defined.
[0005] Several major problems exist with current diabetes risk assessment models: First, accuracy is low, especially when using non-invasive physical examination data, the predictive performance of existing models is insufficient to meet the needs of large-scale applications; second, application scenarios are limited, many models are only applicable to specific datasets or specific populations, and cannot be widely applied in different real-world scenarios; third, personalization is insufficient, existing tools lack the ability to tailor medical advice based on individual differences, and cannot meet the needs of personalized health management; fourth, data integration and processing are insufficient, existing models fail to effectively integrate data from multiple sources (such as lifestyle habits, past medical history, TCM constitution, etc.), resulting in an incomplete and inaccurate assessment.
[0006] Therefore, at this stage, it is necessary to design a hierarchical diabetes risk assessment method, system, and storage medium based on big data and deep learning to solve the above problems. Summary of the Invention
[0007] The purpose of this invention is to provide a hierarchical diabetes risk assessment method, system, and storage medium based on big data and deep learning, to solve the technical problems existing in the prior art. By using large-scale physical examination data and combining artificial intelligence and deep learning technologies, a hierarchical and scenario-based diabetes risk assessment system is constructed. This system can provide high-precision, personalized diabetes risk assessments based on different application scenarios and individual health data, thereby achieving early warning and precise health management.
[0008] To achieve the above objectives, the technical solution of the present invention is as follows:
[0009] A stratified diabetes risk assessment method based on big data and deep learning includes the following steps:
[0010] S1. By combining one-dimensional convolutional neural networks and attention mechanisms in deep learning, and performing hierarchical analysis on physical examination data in different scenarios, the accuracy of diabetes risk assessment can be achieved.
[0011] S2. A normal distribution mapping model is proposed to remap the prediction results to the standard normal distribution, thereby realizing the risk level classification.
[0012] S3. A data perturbation-based method is used to rank the importance of single-sample features for personalized diabetes risk assessment based on individual health conditions.
[0013] Furthermore, in step S1,
[0014] A feature extraction framework based on a one-dimensional convolutional neural network is proposed. All convolutional layers use a kernel size of 3, and a ReLU activation function is added after the last three convolutional operations. The extracted features are flattened into one-dimensional features and then passed to a fully connected neural network for final classification. An attention mechanism is introduced to assign specific weights to the samples.
[0015] First, the input data Perform four layers of one-dimensional convolution operations, where L is the length of the input sequence and C is the number of channels in the input signal; the output of each convolution layer is calculated using the following formula:
[0016] X (l+1) =ReLU(W (l) *X (l) +b (l) )
[0017] Among them, X (l) For the input of the l-th layer, b is the convolution kernel of the l-th layer. (l) Here, k is the kernel size, * denotes the convolution operation, and ReLU(·) is the ReLU activation function. The features extracted through four layers of one-dimensional convolution are denoted as X. (4) ;
[0018] Extracted features X (4) First, the features are flattened into one-dimensional features, then input into a fully connected neural network, introducing an attention mechanism; the features are first mapped to one-dimensional space through a fully connected layer.
[0019] a = Sigmoid(W) a X (4) +b a )
[0020] Among them, W a It is the weight matrix of the attention layer, b a It is a bias term;
[0021] By mapping, the weights a∈[0,1] of the samples are obtained, and these weights are assigned to the input features X′:
[0022] X′'=a⊙X′
[0023] Then, the data is processed through a fully connected layer, during which Dropout and ReLU activation functions are added. Finally, the output layer uses the Sigmoid function to map the result to the range of 0-1, obtaining the final classification result: y = σ(W o X′+b o ).
[0024] Furthermore, in step S2,
[0025] First, generate a set of data, Y, that follows a standard normal distribution and has the same number of samples as the training set; let the number of samples in the training set be N, and then... Y is formed by drawing N samples from the middle, that is
[0026]
[0027] Next, sort Y in ascending order to obtain the sorted target variable Y. sorted Simultaneously, the predicted values X = {x1, x2, ..., x...} from the training set are... N Also sort them in ascending order to get X. sorted ;
[0028] Next, using the sorted X sorted and Y sorted The data is used to train a fully connected neural network; the input is the sorted predicted values X. sorted The output is the sorted target value Y. sorted The mean squared error loss function (MSELoss) is used to optimize the model parameters. The loss function is defined as follows:
[0029]
[0030] Where, f(x) i ) represents the neural network's response to input x i The predicted output, y i For the corresponding target value;
[0031] Based on the model output of the normal distribution mapping, the thresholds for five risk levels are determined by quantiles; the 20%, 40%, 60%, and 80% quantiles are selected as the thresholds for dividing different risk levels; the samples are divided into five levels: very low risk, low risk, medium risk, high risk, and very high risk.
[0032] Let the model predict the output as X = {x1, x2, ..., x...} N}, where N is the number of samples; first, the predicted values are sorted, and the 20%, 40%, 60%, and 80% quantile values are calculated, denoted as P, respectively. 20 P 40 P 60 P 80 Based on these quantiles, the risk level classification rules are as follows: Extremely low risk: when x <P 20 That is, the predicted value of the sample is less than the 20th percentile, indicating low risk: when P 20 ≤x <P 40 That is, the predicted value of the sample is between the 20% and 40% quantiles, which is of medium risk: when P 40 ≤x <P 60 That is, the predicted value of the sample is between the 40% and 60% percentiles, indicating a high risk: when P 60 ≤x <P 80 That is, the predicted value of the sample is between the 60% and 80% percentiles, which is extremely high risk: when x ≥ P 80This means the predicted value of the sample is greater than the 80th percentile; the risk level classification is expressed as:
[0033]
[0034] Furthermore, in step S3,
[0035] A single-sample-based feature contribution ranking method is used to evaluate the relative contribution of each feature to the model's prediction results. Assuming that the predicted value of a given sample1 is X, the contribution of each feature to the prediction results is estimated by calculating the change in the predicted value after random perturbation of each feature.
[0036] For each feature f in sample 1 i Perform n random perturbation experiments; at each perturbation, the eigenvalue f i Change to a random value in:
[0037]
[0038] Where, min i and max i These are the minimum and maximum values of the feature in the entire dataset, respectively;
[0039] Each perturbation feature f i Then, record the new predicted value x output by the model. i,j Where j = 1, 2, ..., n represents the j-th perturbation experiment, and then the characteristic f is calculated. i average predicted value
[0040]
[0041] Each feature f i Contribution value C i By comparing the initial predicted value X with the average predicted value The absolute value of the difference is calculated as follows:
[0042]
[0043] Finally, according to C i Sort the values from largest to smallest to obtain the ranking of the contribution values of each feature.
[0044] The hierarchical diabetes risk assessment system based on big data and deep learning uses the aforementioned hierarchical diabetes risk assessment method based on big data and deep learning to conduct hierarchical diabetes risk assessment.
[0045] A storage medium storing a computer program that, when run, executes the hierarchical diabetes risk assessment method based on big data and deep learning as described above.
[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0047] This invention constructs a hierarchical and scenario-based diabetes risk assessment system based on large-scale physical examination data and combined with artificial intelligence and deep learning technologies. It can provide high-precision, personalized diabetes risk assessments according to different application scenarios and individual health data, thereby achieving early warning and precise health management. This technical solution provides a more scientific and practical auxiliary solution for the early diagnosis and prevention of diabetes. Attached Figure Description
[0048] Figure 1 This is a flowchart illustrating the steps of a hierarchical diabetes risk assessment method based on big data and deep learning. Detailed Implementation
[0049] A stratified diabetes risk assessment method based on big data and deep learning includes the following steps:
[0050] S1. By combining one-dimensional convolutional neural networks and attention mechanisms in deep learning, and performing hierarchical analysis on physical examination data in different scenarios, the accuracy of diabetes risk assessment can be achieved.
[0051] S2. A normal distribution mapping model is proposed to remap the prediction results to the standard normal distribution, thereby realizing the risk level classification.
[0052] S3. A data perturbation-based method is used to rank the importance of single-sample features for personalized diabetes risk assessment based on individual health conditions.
[0053] (1) Data
[0054] First, we excluded data columns with more than 50% missing values and removed features that had no significant discriminative effect in the diagnosis of coronary heart disease, such as cervical smears and self-assessments of health status in the elderly. Furthermore, we cleaned up non-standard items in the data, such as correcting non-standard suffixes in the occupation column and converting certain medical terms into numerical forms so that the model could better handle this information. We also converted multi-select features such as cerebrovascular diseases into one-hot encoding. After these steps, we finally obtained 569,634 data points of healthy individuals (negative samples) containing 131 features and 157,453 data points of coronary heart disease patients (positive samples) containing 131 features, with a positive-to-negative sample ratio of approximately 1:3. Next, we divided the data into training and test sets in a 9:1 ratio, and then further divided the training set into the final training and validation sets in a 9:1 ratio.
[0055] (2) Method
[0056] We first divided the physical examination data into five versions: the first version includes resident information, routine physical examination information, blood routine information, and biochemical indicator information; the second version, based on the first version, adds lifestyle information and other examination results; the third version further adds other disease information to the second version; the fourth version, based on the third version, adds traditional Chinese medicine constitution examination information; and the fifth version is a special version, namely the non-invasive self-test version, which is designed for home testing groups, meaning that no new data needs to be obtained from the hospital, only old illnesses and routine physical information are needed to complete the diagnostic prediction. Our model is based on a one-dimensional convolutional architecture and optimized with an attention mechanism. In addition, we also constructed a normal distribution mapping model to remap the prediction results of the ensemble model onto a normal distribution. Based on the remapped data, we used data at the 20%, 40%, 60%, and 80% positions as thresholds to classify the prediction results into five risk levels: below 20% is low risk, 20%-40% is relatively low risk, 40%-60% is moderate risk, 60%-80% is relatively high risk, and above 80% is high risk. Finally, we rank the importance of single-sample features based on a data perturbation method.
[0057] (2.1) Data version division
[0058] The first version of the data features includes resident information, routine physical examination information, blood routine information, and biochemical indicator information. Resident information includes age and gender. Routine physical examination information includes body temperature, pulse rate, heart rate, respiratory rate, height, weight, waist circumference, body mass index, mean systolic blood pressure, mean diastolic blood pressure, pulse pressure difference, waist-to-weight ratio, waist circumference-to-body mass ratio, and waist-to-height ratio. Blood routine information includes hemoglobin, white blood cells, and platelets. Biochemical indicator information includes fasting blood glucose, serum alanine aminotransferase, serum aspartate aminotransferase, total bilirubin, serum creatinine, blood urea nitrogen, total cholesterol, triglycerides, serum low-density lipoprotein cholesterol, serum high-density lipoprotein cholesterol, urine protein, urine glucose, urine ketones, and urine occult blood.
[0059] The second version adds lifestyle information and other examination information to the first version. The lifestyle information includes exercise frequency, diet, smoking status, drinking frequency and whether there has been intoxication in the past year. Other examination information includes electrocardiogram, other B-ultrasound, B-ultrasound, and chest X-ray.
[0060] The third version adds information on other diseases to the second version, including dorsalis pedis artery pulsation, cerebrovascular disease, kidney disease, heart disease, vascular disease, eye disease, nervous system disease, other systemic diseases and symptoms.
[0061] The fourth version adds TCM constitution examination information to the third version. The TCM constitution examination information includes balanced constitution, qi deficiency constitution, yang deficiency constitution, yin deficiency constitution, phlegm-dampness constitution, damp-heat constitution, blood stasis constitution, qi stagnation constitution, and special constitution.
[0062] The fifth version is a non-invasive self-test version, which includes resident information, general physical examination information, lifestyle-related information, and other disease information. Resident information includes age, gender, ethnicity, blood type, marital status, type of residence, education level, whether it is a migrant population, whether it is a poverty-stricken population, occupation code, and household registration type. General physical examination information includes height, weight, waist circumference, body mass index, waist-to-weight ratio, waist-to-body ratio, and waist-to-height ratio. Lifestyle-related information includes exercise frequency, dietary habits, smoking status, drinking frequency, and whether there has been intoxication in the past year. Other disease information includes cerebrovascular diseases, kidney diseases, heart diseases, vascular diseases, eye diseases, nervous system diseases, other system diseases, and symptoms.
[0063] (2.2) Diabetes Prediction Model
[0064] In this study, we propose a feature extraction framework based on a one-dimensional convolutional neural network (CNN) for diabetes diagnosis. All convolutional layers use a kernel size of 3, and a ReLU activation function is added after the last three convolutional operations. The extracted features are flattened into one-dimensional features and then passed to a fully connected neural network for final classification. This network introduces an attention mechanism to assign specific weights to samples.
[0065] First, the input data A four-layer one-dimensional convolution operation is performed, where L is the length of the input sequence and C is the number of channels in the input signal. The output of each convolution layer is calculated using the following formula:
[0066] X (l+1) =ReLU(W (l) *X (l) +b (l) )
[0067] Among them, X (l) For the input of the l-th layer, b is the convolution kernel of the l-th layer. (l) Here, k is the kernel size (k = 3 in this model), * denotes the convolution operation, and ReLU(·) is the ReLU activation function. The features extracted through four layers of one-dimensional convolution are denoted as X. (4) .
[0068] Extracted features X (4) First, the features are flattened into one-dimensional features, then input into a fully connected neural network that incorporates an attention mechanism. The features are first mapped to one-dimensional space through a fully connected layer.
[0069] a = Sigmoid(W) a X (4) +b a )
[0070] Among them, W a It is the weight matrix of the attention layer, b a It is a bias term.
[0071] Through this mapping, we obtain the weights a∈[0,1] of the samples and assign them to the input features X′:
[0072] X′'=a⊙X′
[0073] Then, after further processing by the fully connected layer, we incorporate Dropout (with a weight of 0.1) and the ReLU activation function. Finally, the output layer uses the Sigmoid function to map the result to the range of 0-1, yielding the final classification result.
[0074] y = σ(Wo X′+b o )
[0075] (2.3) Normal distribution mapping model
[0076] First, we generated a set of data, Y, with the same number of samples as the training set, following a standard normal distribution. Let the number of samples in the training set be N, and we use a standard normal distribution... Y is formed by drawing N samples from the middle, that is
[0077]
[0078] Next, we sort Y in ascending order to obtain the sorted target variable Y. sorted Simultaneously, the predicted values X = {x1, x2, ..., x...} from the training set are... N Also sort them in ascending order to get X. sorted。
[0079] Next, using the sorted X sorted and Y sorted The data is used to train a fully connected neural network. Specifically, the input is the sorted predicted values X. sorted The output is the sorted target value Y. sorted We use the mean squared error loss function (MSELoss) to optimize the model parameters. The loss function is defined as follows:
[0080]
[0081] Where, f(x) i ) represents the neural network's response to input x i The predicted output, y i This is the corresponding target value.
[0082] (2.4) Risk Classification
[0083] In this study, we determined the thresholds for five risk levels based on the model output of the normal distribution mapping using quantiles. Specifically, we selected the 20%, 40%, 60%, and 80% quantiles as the thresholds for different risk levels. Using these thresholds, the samples were divided into five levels: very low risk, low risk, moderate risk, high risk, and very high risk.
[0084] Let the model predict the output as X = {x1, x2, ..., x...} N}, where N is the sample size. To determine the thresholds for different risk levels, we first sort the predicted values and calculate the 20%, 40%, 60%, and 80% quantile values, denoted as P0. 20 P 40 P60 P 80 Based on these quantiles, the risk level classification rules are as follows: Extremely low risk: when x <P 20 That is, the predicted value of the sample is less than the 20th percentile, indicating low risk: when P 20 ≤x <P 40 That is, the predicted value of the sample is between the 20% and 40% quantiles, which is of medium risk: when P 40 ≤x <P 60 That is, the predicted value of the sample is between the 40% and 60% percentiles, indicating a high risk: when P 60 ≤x <P 80 That is, the predicted value of the sample is between the 60% and 80% percentiles, which is extremely high risk: when x ≥ P 80 This means the predicted value of the sample is greater than the 80th percentile. Therefore, the risk level classification can be expressed as:
[0085]
[0086] (2.5) Single Sample Feature Ranking
[0087] In this study, we used a single-sample feature contribution ranking method to evaluate the relative contribution of each feature to the model's prediction results. Assume the predicted value of a given sample1 is X. By randomly perturbing each feature, we calculate the change in the predicted value after the perturbation to estimate the feature's contribution to the prediction result.
[0088] For each feature f in sample 1 i Perform n random perturbation experiments. At each perturbation, the eigenvalue f is... i Change to a random value in:
[0089]
[0090] Where, min i and max i These are the minimum and maximum values of the feature in the entire dataset, respectively.
[0091] Each perturbation feature f i Then, record the new predicted value x output by the model. i,j (where j = 1, 2, ..., n represents the j-th perturbation experiment), and then the characteristic f is calculated. i average predicted value
[0092]
[0093] Each feature f i Contribution value Ci By comparing the initial predicted value X with the average predicted value The absolute value of the difference is calculated as follows:
[0094]
[0095] Finally, according to C i The values are sorted from largest to smallest to obtain a ranking of the contribution values of each feature. The larger the contribution value, the more significant the impact of that feature on the model's predicted value.
Claims
1. A hierarchical diabetes risk assessment method based on big data and deep learning, characterized in that, Includes the following steps: S1. By combining one-dimensional convolutional neural networks and attention mechanisms in deep learning, and performing hierarchical analysis on physical examination data in different scenarios, the accuracy of diabetes risk assessment can be achieved. S2. A normal distribution mapping model is proposed to remap the prediction results to the standard normal distribution, thereby realizing the risk level classification. S3. Use a data perturbation-based method to rank the importance of single-sample features and conduct personalized diabetes risk assessments for individual health conditions. In step S1, A feature extraction framework based on a one-dimensional convolutional neural network is proposed. All convolutional layers use a kernel size of 3, and a ReLU activation function is added after the last three convolutional operations. The extracted features are flattened into one-dimensional features and then passed to a fully connected neural network for final classification. An attention mechanism is introduced to assign specific weights to the samples. First, the input data Perform four layers of one-dimensional convolution operations, where The length of the input sequence. The number of channels in the input signal; the output of each convolutional layer is calculated using the following formula: in, For the first Layer input, For the first The convolution kernel of the layer, For bias terms, The size of the convolution kernel, sign This represents the convolution operation. for Activation function; the features extracted through four layers of one-dimensional convolution are denoted as... ; Extracted features First, the features are flattened into one-dimensional features, then input into a fully connected neural network, introducing an attention mechanism; the features are first mapped to one-dimensional space through a fully connected layer. in, It is the weight matrix of the attention layer. It is a bias term; The weights of the samples are obtained through mapping. and assign it to the input features : Then, the data is processed through a fully connected layer, during which Dropout and ReLU activation functions are added; finally, the Sigmoid function is used in the output layer to map the result to the range of 0-1, yielding the final classification result. ; In step S2, First, generate a set of data that follows a standard normal distribution and has the same number of samples as the training set as the target variable. Let the number of training set samples be... From the standard normal distribution Extraction Each sample constitutes ,Right now Next, regarding Sort the variables from smallest to largest to obtain the sorted target variables. Simultaneously, the predicted values from the training set... Also sorted in ascending order, we get Next, using the sorted... and The data is used to train a fully connected neural network; the input is the sorted predicted values. The output is the sorted target value. The mean squared error loss function MSELoss is used to optimize the model parameters. The loss function is defined as follows: in, This indicates that the neural network responds to the input. The predicted output, For the corresponding target value; Based on the model output of the normal distribution mapping, the thresholds for five risk levels are determined by quantiles; the 20%, 40%, 60%, and 80% quantiles are selected as the thresholds for dividing different risk levels; the samples are divided into five levels: low risk, lower risk, medium risk, higher risk, and high risk. Let the model predict the output as follows: ,in Given the sample size; first, sort the predicted values and calculate the 20%, 40%, 60%, and 80% quantiles, denoted as _____. Based on these quantiles, the risk level classification rules are as follows: Low risk: When That is, the predicted value of the sample is less than quantile, lower risk: when That is, the predicted value of the sample is located in Between quantiles, moderate risk: when That is, the predicted value of the sample is located in Between quantiles, higher risk: when That is, the predicted value of the sample is located in Between quantiles, high risk: when That is, the predicted value of the sample is greater than Quantiles; risk levels are represented as follows: 。 2. The hierarchical diabetes risk assessment method based on big data and deep learning according to claim 1, characterized in that, In step S3, A single-sample feature contribution ranking method is used to evaluate the relative contribution of each feature to the model's prediction results; assuming the predicted value of a given sample1 is... By randomly perturbing each feature, the change in the predicted value after the perturbation is calculated to estimate the contribution of that feature to the prediction result. For each feature in sample 1 ,implement This is a random perturbation experiment; at each perturbation, the eigenvalues are... Change to a random value ,in: in, and These are the minimum and maximum values of the feature in the entire dataset, respectively; Features of each perturbation Then, record the new predicted values output by the model. ,in Indicates the first Secondary perturbation experiment, then calculate the characteristics average predicted value : Each feature Contribution value Based on the initial predicted value Compared with the average forecast value The absolute value of the difference is calculated as follows: Finally, according to Sort the values from largest to smallest to obtain the ranking of the contribution values of each feature.
3. A hierarchical diabetes risk assessment system based on big data and deep learning, characterized in that: The stratified diabetes risk assessment method based on big data and deep learning as described in any one of claims 1-2 is used to conduct the stratified diabetes risk assessment.
4. A storage medium, characterized in that, The storage medium stores a computer program, which, when run, executes the hierarchical diabetes risk assessment method based on big data and deep learning as described in any one of claims 1-2.
