Intelligent risk prediction method based on neural network

Abstract

The invention relates to the technical field of disease prediction, in particular to an intelligent risk prediction method based on a neural network, and the method comprises the steps: collecting blood glucose and body mass index data, carrying out the clustering analysis, and constructing a multi-risk group feature; a risk transfer track before attack is recognized by combining a historical index change path and time sequence analysis; and calculating an individual risk coefficient by using deep learning, and comparing the individual risk coefficient with a risk threshold to generate early warning information, thereby realizing dynamic accurate evaluation. According to the method, by collecting the blood glucose level and body weight index data and conducting standardization processing, risk group division based on health index clustering can be achieved, a group recognition system with health feature differentiation is constructed, and the structured understanding of the individual health trend is enhanced. And on the basis of the clustered group, extracting a fluctuation path of individual historical indexes by means of a time sequence mode, identifying key transfer characteristics before disease attack, and enhancing the traceability of the disease formation process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of disease prediction, and in particular to an intelligent risk prediction method based on a neural network. BACKGROUND

[0002] The technical field of the intelligent risk prediction method includes the field of disease prediction. The field of disease prediction technology is committed to predicting the risk of individuals or groups of developing certain diseases through various data analysis and pattern recognition means. The core of this field is a data-driven model that uses patient health data, genetic information, lifestyle habits, and other multi-dimensional information, combined with statistical methods, machine learning algorithms, and other methods, to assess the risk of disease occurrence.

[0003] Among them, the intelligent risk prediction method refers to a technical solution for predicting the health risks of individuals or groups through intelligent technology. It includes the analysis of multiple factors affecting the risk of disease occurrence, including statistical health record data, laboratory test results, imaging data, and lifestyle information.

[0004] In the past intelligent risk prediction technology, although it covers a variety of sources of health information, it is weak in identifying the heterogeneity of the health status within the group, resulting in a lack of targetedness in predicting results for multiple types of patients. Risk identification is often based on linear analysis of static data, ignoring the dynamic changes of key indicators in the early stages of disease, making it difficult to accurately capture the risk transfer trend and reducing the sensitivity of early warning signals before the onset. In addition, the depth of the use of individual historical health data is insufficient, and its potential in risk evolution path identification has not been tapped, resulting in weak ability to identify early disease evolution patterns. In terms of risk judgment mechanism, the common approach is based on current single-point indicators, lacking analysis methods that combine historical fluctuation trajectories, and easily ignoring the potential nonlinear development patterns of the disease. At the early warning mechanism level, the update mechanism relies on manual intervention or periodic determination, and fails to establish a continuous dynamic feedback process, resulting in a lag in risk state response. For example, when a patient's body mass index slowly increases and blood sugar fluctuates slightly, a static model often fails to identify this small but continuous trend, delaying the intervention opportunity and affecting the effectiveness of health management. The existing technology has not formed an integrated linkage of data collection, risk identification, and dynamic feedback in the overall structure, and the prediction and intervention effectiveness for complex chronic diseases is limited. SUMMARY

[0005] The purpose of the present application is to solve the problems existing in the prior art and to propose an intelligent risk prediction method based on a neural network.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: an intelligent risk prediction method based on a neural network, comprising the following steps: S1: Collect blood glucose levels and body mass index of diabetes patients, use unsupervised learning algorithm to cluster the standardized data, divide multiple risk groups according to blood glucose and BMI, identify the differences in health characteristics of each group, and generate diabetes risk analysis results; S2: Collect the historical blood glucose levels and body mass index change path of the diabetes patient group in the specified period before the onset in the diabetes risk analysis results, use time series analysis to calculate the risk transfer index, identify the typical risk transfer path before the onset, and generate the diabetes transfer trajectory analysis results; S3: Use the diabetes risk analysis results as the input of the neural network model for deep learning, combine the historical blood glucose levels and body mass index change path in the diabetes transfer trajectory analysis results, and calculate the corresponding diabetes risk coefficient; S4: Compare the diabetes risk coefficient with the preset risk threshold, evaluate the risk state and potential disease deterioration of the diabetes patient, and generate diabetes risk warning information.

[0007] The diabetes risk analysis results include blood glucose levels, body mass index, health group division, and potential risk patterns; the diabetes transfer trajectory analysis results include risk transfer index, pre-onset blood glucose fluctuation, pre-onset body mass index fluctuation, typical risk transfer path, and physiological changes that trigger diabetes; the diabetes risk coefficient includes current blood glucose level, current body mass index, historical blood glucose level change, historical body mass index change, and fluctuation comparison; and the diabetes risk warning information includes current risk state, potential disease deterioration risk, complication risk, and health management warning signal.

[0008] The diabetes risk analysis results include blood glucose levels, body mass index, health group division, and potential risk patterns; the diabetes transfer trajectory analysis results include risk transfer index, pre-onset blood glucose fluctuation, pre-onset body mass index fluctuation, typical risk transfer path, and physiological changes that trigger diabetes; the diabetes risk coefficient includes current blood glucose level, current body mass index, historical blood glucose level change, historical body mass index change, and fluctuation comparison; and the diabetes risk warning information includes current risk state, potential disease deterioration risk, complication risk, and health management warning signal. S101: Collect blood glucose levels and body mass index data of diabetes patients, standardize the data, and based on the standardized data, cluster the blood glucose levels and body mass index of each diabetes patient, obtain the center point, mean, and variance statistics of the cluster data distribution, and generate diabetes health group statistical information; S102: According to the diabetes health group statistical information, use unsupervised learning algorithm to analyze the clustered patient data, divide the patients into multiple groups according to the distribution of blood glucose levels and body mass index, each group represents a potential diabetes risk analysis result, and generate diabetes risk group division results; S103: Based on the diabetes risk group division results, the statistical characteristic values ​​of each group are compared with the risk factors for diabetes, the differences in health characteristics exhibited by the patient groups during the diabetes development process are identified, and the diabetes risk analysis results are generated.

[0009] The present invention has been improved in that the specific steps of collecting the historical blood glucose level and body mass index change paths of the diabetic patient group in a specified period before the onset of the disease in the diabetes risk analysis results, calculating the risk transfer index using time series analysis, identifying the typical risk transfer paths before the onset of the disease, and generating the diabetes transfer trajectory analysis results are as follows: S201: Collecting historical data on blood glucose levels and body mass index of a group of diabetic patients in the diabetes risk analysis results for a specified period before the onset of the disease, arranging the data in time series, and generating historical blood glucose and body mass index change data by arranging and converting the data; S202: Based on the historical blood glucose and body mass index change data, a time series analysis method is used to calculate a risk transfer index, and by comparing the fluctuations of the patient's blood glucose and body mass index over a specified period, key fluctuation characteristics are extracted to obtain a fluctuation characteristic analysis result; S203: Based on the fluctuation characteristic analysis results, identify the typical risk transfer path of each diabetic patient before the onset of the disease, determine the relationship between the fluctuation pattern of health data and diabetes risk, generate diabetes transfer trajectory analysis results, and identify the key physiological changes that cause diabetes.

[0010] The present invention has the following improvements: for calculating the risk transfer index , using the formula: ; in, and is the adjustment factor, , is the baseline value of blood sugar changes, is the baseline value of body mass index change, represents the total number of time points in the analysis period, and represent the difference between consecutive time points, Indicates a point in time.

[0011] The present invention has been improved in that the diabetes risk analysis results are used as input to a neural network model for deep learning, and the specific steps for calculating the corresponding diabetes risk coefficient are as follows, combining the historical blood glucose level and body mass index change paths in the diabetes transfer trajectory analysis results: S301: Integrate the diabetes risk analysis result and the diabetes transfer trajectory analysis result, extract the current blood glucose level and body mass index data of the diabetes patient, and the historical blood glucose and body mass index change path of each patient and pre-process to generate standardized input data; S302: Input the standardized input data into the neural network for training, and the neural network will combine the current blood glucose level, body mass index and historical change path to learn the health pattern of the patient, calculate the diabetes risk coefficient of each patient, and quantify the diabetes risk of the patient in the current health state.

[0012] The present application improves that for calculating the diabetes risk coefficient of each patient , the formula is: ; Wherein, represents the blood glucose at the time point, represents the body mass index at the time point, and are the adjustment factors at each time point, is a Sigmoid activation function, is a bias term, is the number of continuous data points collected in the time window.

[0013] The present application improves that the diabetes risk coefficient is compared with the preset risk threshold to evaluate the risk state and potential disease deterioration of the diabetes patient, and the specific steps of generating the diabetes risk warning information are as follows: S401: Compare the diabetes risk coefficient with the preset risk threshold to determine whether the patient is in a risk state, and generate a risk state judgment result according to the comparison result; S402: According to the risk state judgment result, evaluate the potential disease deterioration or complication risk of the patient, identify the physiological changes that lead to the aggravation of diabetes symptoms, analyze the risk development trend of the patient by comparing the current blood glucose level and body mass index, and generate diabetes risk warning information.

[0014] The present application improves that it further includes step S5, based on the diabetes risk warning information, real-time updating the risk state of the diabetes patient, and generating an updated risk assessment result; The risk assessment result includes the updated risk state, the latest health data evaluation result, and the risk level adjustment.

[0015] The present application improves that based on the diabetes risk warning information, the risk state of the diabetes patient is real-time updated, and the specific steps of generating an updated risk assessment result are as follows: S501: Based on the diabetes risk warning information, obtain regular physical examination data of the diabetic patient and monitor and record it, including the latest health indicators of blood sugar level and body mass index, to generate regular physical examination health data; S502: Based on the regular physical examination health data, re-evaluate the risk status of the diabetic patient, compare the updated health indicators with the previous diabetes risk coefficient, determine whether the patient's health status has changed, and adjust the diabetes risk status according to the changes to generate an updated risk assessment result.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are: In this invention, by collecting and standardizing blood glucose and body mass index data, risk groups can be identified based on health indicator clustering, building a group identification system with differentiated health characteristics and enhancing a structured understanding of individual health trends. Based on the clustered groups, a time series approach is used to extract the fluctuation paths of individual historical indicators, identifying key transition features before onset and strengthening the ability to trace the disease development process. The calculation of the risk transfer index incorporates the degree of change and fluctuation comparison factors between time points to quantify the underlying risk transmission relationship of indicator changes and improve the accuracy of identifying disease evolution paths. Historical indicator fluctuations and current health status are simultaneously incorporated into the neural network training process, ensuring that the model considers both cross-sectional information and longitudinal evolution trends, incorporating dynamic and individual factors into risk prediction. The calculation of the risk coefficient establishes a precise measurement method for individual health changes by integrating multidimensional indicators, improving the granularity of risk assessment. The assessment results are compared with the risk coefficient threshold, supplemented by real-time data backtracking and trend analysis, to output detailed early warning information, which can effectively identify potential signs of deterioration and promote timely intervention. Risk status is dynamically updated based on early warning signals, and regular physical examination data is fed back into the assessment mechanism, creating a closed-loop feedback system that continuously tracks patient health status and adjusts risk levels in real time. The overall process integrates data clustering, time series analysis, deep learning modeling, and real-time feedback to establish a multi-stage, multi-angle, and dynamically evolving diabetes risk prediction system, significantly enhancing the forward-looking and personalized responsiveness of health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of the method of the present invention; Figure 2 This is a detailed flow chart of step S1 of the present invention; Figure 3 This is a schematic diagram of a detailed process of step S2 of the present invention; Figure 4 This is a detailed flow chart of step S3 of the present invention; Figure 5The step S4 of the application is refined in the flowchart. Figure 6 The step S5 of the application is refined in the flowchart. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0019] In the description of the present application, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate the orientation or positional relationship shown in the drawings, which is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application. In addition, in the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly and specifically limited.

[0020] Please refer to Figure 1 The present application provides a technical scheme: an intelligent risk prediction method based on a neural network, comprising the following steps: S1: collecting blood glucose levels and body mass indexes of diabetic patients, performing cluster analysis on the standardized blood glucose levels and body mass indexes through an unsupervised learning algorithm, dividing each diabetic patient into multiple populations according to blood glucose levels and body mass indexes, each population representing a potential diabetes risk pattern, identifying the health feature differences of each diabetic patient population in the diabetes development process, and generating a diabetes risk analysis result; S2: collecting the historical blood glucose levels and body mass index change paths of the diabetic patient population in the diabetes risk pattern at a specified period before the onset, calculating the risk transfer index using a time series analysis method, identifying the typical risk transfer path of each diabetic patient before the onset according to the risk transfer index, judging the physiological changes that trigger the diabetes risk, and generating a diabetes transfer trajectory analysis result; S3: using the diabetes risk analysis result as the input of a neural network model for deep learning, combining the current blood glucose levels and body mass indexes of the diabetic patient with the historical blood glucose levels and body mass index change paths in the diabetes transfer trajectory analysis result, comparing the current health data of the patient with the data fluctuations in the historical transfer path, and calculating the corresponding diabetes risk coefficient; S4: comparing the diabetes risk coefficient with the preset risk threshold, determining the current risk state of the diabetes patient according to the comparison result, evaluating the potential disease exacerbation or complication risk of the existing risk diabetes patient, providing a warning signal for the health management of the patient, generating diabetes risk warning information; The diabetes risk analysis result includes blood glucose level, body mass index, health group division, and potential risk mode; the diabetes transfer trajectory analysis result includes risk transfer index, pre-onset blood glucose fluctuation, pre-onset body mass index fluctuation, typical risk transfer path, and physiological change triggering diabetes; the diabetes risk coefficient includes current blood glucose level, current body mass index, historical blood glucose level change, historical body mass index change, and fluctuation comparison; and the diabetes risk warning information includes current risk state, potential disease exacerbation risk, complication risk, and health management warning signal.

[0021] Please refer to Figure 2 , collect the blood glucose level and body mass index of the diabetes patient, use unsupervised learning algorithm to cluster the standardized data, divide multiple risk groups according to blood glucose and BMI, identify the health feature differences of each group, and the specific steps of generating the diabetes risk analysis result are as follows: S101: Collect the blood glucose level and body mass index data of the diabetes patient, standardize the data, cluster the blood glucose level and body mass index of each diabetes patient based on the standardized data, obtain the center point, mean, and variance statistics of the cluster data distribution, and generate diabetes health group statistical information; First, the patient's blood glucose level and body mass index data need to be collected regularly. Assuming that each patient has a blood glucose level measurement and a body mass index check once a month, after collecting these data, the first step is to standardize the data, which includes converting the data to a relative standard numerical range. A commonly used method is to subtract the mean of the data from each data point and divide by the standard deviation. The formula is as follows: standardized data = (original data - mean) / standard deviation. For example, assuming a patient's blood glucose level is 150 mg / dL, the standard deviation is 30, and the mean is 120, the standardized data is: (150-120) / 30 = 1.00. The standardized blood glucose level and body mass index can be easily clustered to reduce the influence of units and dimensions. Next, based on the standardized data, clustering analysis is performed to divide all diabetic patients into different groups according to the similarity of blood glucose level and body mass index. Unsupervised learning algorithms such as K-means clustering are used. The algorithm will cluster the patients into different groups according to their blood glucose and body mass index distribution. By calculating the mean and variance of each group of data points, the center point and statistical quantity of the clustered data can be obtained, further describing the characteristics of each group. For example, if a patient group has a mean blood glucose level of 130 and a mean body mass index of 27, this group represents a relatively high-risk diabetic group. Finally, statistical information of the diabetic health group is generated based on these statistics to provide data support for subsequent risk assessment.

[0022] S102: According to the statistical information of the diabetic health group, use unsupervised learning algorithm to analyze the clustered patient data, divide the patients into multiple groups according to the distribution of blood glucose level and body mass index, each group represents a potential diabetes risk analysis result, generate diabetes risk group division result; The unsupervised learning algorithm further analyzes the distribution of blood glucose levels and body mass index for each group to identify potential health risk differences among the diabetic patient groups. Assuming that some groups with a more concentrated distribution in the diabetic patient group may be related to well-controlled diabetic patients, while groups with a more extensive distribution may reflect poor disease control or potential complications. In Python, using the KMeans class of the scikit-learn library, the data can be clustered and the group label of each data point can be obtained by the fit_predict method. By analyzing the distribution of patient groups, the risk level of each group can be obtained. For example, in a certain diabetic patient group, the average blood glucose level is 180 and the body mass index is 32, while in another group, the average blood glucose level is 120 and the body mass index is 25, which shows different disease risk levels. The analysis results show that the higher the blood glucose and body mass index, the higher the risk of complications in patients. Therefore, based on these statistical information, the purpose of data clustering is to divide patients into multiple groups according to different risk levels, and each group represents a potential diabetes risk pattern. In this way, patients are assigned to the corresponding group for subsequent risk assessment. Finally, through these analyses, the diabetes risk group division results are generated, and the health risk level of each group is calibrated.

[0023] S103: Based on the diabetes risk group division results, compare the statistical characteristic values of each group with the risk factors of diabetes occurrence to identify the health characteristic differences exhibited by the patient groups in the diabetes occurrence process, and generate diabetes risk analysis results; By comparing the statistical characteristic values of each group with the risk factors of diabetes occurrence, the health characteristic differences of each group in the diabetes occurrence process are further analyzed. Assuming that the blood glucose level of some patient groups is higher than 150mg / dL for a long time, and the body mass index exceeds 30, the risk of diabetes occurrence in these patient groups will significantly increase over time. For example, if a group of patients has a blood glucose level higher than 140mg / dL for the past year and a body mass index above 30, this group is more likely to develop diabetes complications such as retinopathy and cardiovascular disease compared to groups with better blood glucose control. On this basis, by comparing the health data of different groups, the health characteristic differences that may occur in the diabetes occurrence process can be identified. By comparing the health characteristics of each group with the risk factors of diabetes, such as high blood glucose and high body mass index, the risk level of different patient groups can be identified to help clinicians assess the health status of patients. Through this comparative analysis, the diabetes risk analysis results are finally generated, providing personalized diabetes risk assessment information for patients.

[0024] Please refer to Figure 3, collect the historical blood glucose level and body mass index change paths of the diabetic patient group in the specified period before the onset of diabetes risk analysis results, use time series analysis to calculate the risk transfer index, identify the typical risk transfer path before the onset of the disease, and generate the diabetes transfer trajectory analysis results in the following specific steps: S201: Collect historical data on blood glucose levels and body mass index of a group of diabetic patients in a specified period before the onset of the disease in the diabetes risk analysis results, organize the data into time series, and generate historical blood glucose and body mass index change data by organizing and converting the data; First, a clear time period must be specified, for example, using the six months prior to onset as a time window. Then, raw blood glucose and body mass index records for that period are extracted from the individual patient health monitoring data recorded in the electronic medical record system. For example, patient A's blood glucose records at six time points were 132, 140, 148, 155, 150, and 160 mg / dL, and his body mass index records were 27.2, 27.6, 28.1, 28.3, 28.7, and 29.1 kg / m². The acquired data must be organized into a time series data structure in chronological order to meet the requirements of subsequent dynamic feature modeling. To achieve a unified structure, data collation removes missing or irregular data points and interpolates to ensure time series continuity. The conversion process also requires data alignment based on temporal granularity, for example, by dividing all records into monthly intervals with a unified 30-day period to maintain consistency in the analytical foundation, resulting in standardized, time-ordered, and unit-separated health monitoring series data.

[0025] S202: Based on historical blood glucose and body mass index change data, a time series analysis method is used to calculate a risk transfer index. By comparing the fluctuations of the patient's blood glucose and body mass index over a specified period, key fluctuation characteristics are extracted to obtain fluctuation characteristic analysis results. First, a first-order difference sequence is constructed for each indicator in the time series to capture its continuous change rate, which is defined as ; in, Indicates the Blood glucose value at each time point (unit: mg / dL), Indicates the blood glucose value at the previous time point; Indicates the Body mass index at each time point (unit: kg / m²), represents the body mass index at the previous time point, and respectively, representing the difference between consecutive time points, used to describe the fluctuation amplitude of blood glucose and body mass index. To facilitate the assessment of the risk contribution of the two indicators at different scales, the standardized unit change ratio form is adopted, and the risk transfer index is calculated as follows: ; wherein, and are adjustment factors used to balance the influence of blood glucose and body mass index on the overall risk transfer index, satisfying . The setting of the adjustment factor needs to be based on clinical research or statistical regression analysis for parameter calibration, for example, according to previous data analysis, it is found that the explanatory degree of blood glucose fluctuation for the incidence of diabetes is 60%, and the explanatory degree of body mass index change is 40%, based on which , . In addition, is the blood glucose unit change reference value (set as 5 mg / dL), is the body mass index unit change reference value (set as 0.2 kg / m²), represents the total number of time points in the analysis period. Taking a patient as an example, if his blood glucose rises by an average of 6 mg / dL per month and his body mass index rises by an average of 0.3 kg / m² per month within 6 months, then ; According to the experience threshold value, is judged as low risk, is judged as medium risk, is judged as high risk, and then the fluctuation persistence, change amplitude proportion and other characteristics are extracted, and finally the fluctuation characteristic analysis result is obtained.

[0026] S203: Based on the fluctuation characteristic analysis result, identify the typical risk transfer path of each diabetes patient before the onset of diabetes, judge the relationship between the fluctuation mode of health data and the risk of diabetes, generate the diabetes transfer trajectory analysis result, and identify the key physiological changes that trigger diabetes; First, all patients with complete fluctuation characteristics are constructed into a trajectory sample set, and the trajectory path is defined as a chain of continuous changes in the state in the time series ; wherein, represents the state point at the time point, including the blood glucose change value (unit: mg / dL) and the body mass index change value (unit: kg / m2), and the trajectory line is formed by connecting these state points, representing the risk migration path of the individual in the time dimension. Then, based on the trajectory similarity, the density clustering method (such as DBSCAN) is used to aggregate and identify all trajectories, and divided into structural types such as "continuous rising type", "periodic fluctuation type", "fluctuation superposition type", "steady state maintenance type" and so on. For example, if a trajectory meets the condition of "continuous rising type" at 3 consecutive time points 、 , it can be determined as a "continuous rising type" trajectory. Further, the distance between the tail point of the trajectory and the diagnosis time is calculated, and if most of the trajectories present this pattern within 1 month before the onset, the correspondence between the pattern and the onset of diabetes can be established. Finally, the typical trajectory structure is matched with the fluctuation sequence of the new individual, and the trajectory group that meets the risk characteristics of the onset is extracted, generating the diabetes transfer trajectory analysis result, and identifying the structure of "continuous rising of blood glucose + cumulative growth of body mass index" as the potential key physiological change characteristics.

[0027] Please refer to Figure 4 The diabetes risk analysis result is used as the input of the neural network model for deep learning, and the historical blood glucose level and body mass index change path in the diabetes transfer trajectory analysis result are combined to calculate the specific steps of the corresponding diabetes risk coefficient as follows: S301: Integrate the diabetes risk analysis result and the diabetes transfer trajectory analysis result, extract the current blood glucose level and body mass index data of the diabetes patient, and the historical blood glucose and body mass index change path of each patient, and perform preprocessing to generate standardized input data; First, the current blood glucose level and body mass index of each patient are extracted, and the current blood glucose value and body mass index of the patient are obtained, for example, the current blood glucose of patient A is 155 mg / dL, and the body mass index is 29.0 kg / m2. Next, the historical data of the patient is extracted, including the blood glucose change and body mass index change in the past 6 months, which will be used as the historical change path input model. In order to input these data into the neural network for training, the data needs to be standardized to ensure that all input data has the same unit and range. The sequence data of historical blood glucose and body mass index change will be arranged in time series form, including the monthly blood glucose change and body mass index change. All data is preprocessed to eliminate any abnormal or missing values and interpolated. The standardized data will include the change amount of blood glucose and body mass index at each time point and the current blood glucose and body mass index information, which will be used as standardized input data for subsequent neural network training.

[0028] S302: input the standardized input data into the neural network for training, the neural network will combine the current blood glucose level, body mass index and historical change path, learn the health pattern of the patient, calculate the diabetes risk coefficient of each patient, which is used to quantify the diabetes risk of the patient under the current health status; The standardized input data is input into the neural network for training, and the input of the neural network includes the current patient's blood glucose level, body mass index, and its historical blood glucose change and body mass index change data. Since the units of blood glucose and body mass index are different, it is necessary to convert them into a unified standardized format to ensure that they can be input into the network together for processing. For example, the unit of blood glucose is mg / dL, and the unit of body mass index is kg / m², so it is necessary to standardize the blood glucose and body mass index data to make their units consistent and ensure that the influence of different units is reasonably adjusted.

[0029] The structure of the neural network is learned by passing these input features into each layer and gradually learning, and finally outputs the diabetes risk coefficient. The formula of the neural network model is as follows: ; Wherein, represents the calculated diabetes risk coefficient, represents the blood glucose value (standardized) at the time point, represents the body mass index value (standardized) at the time point, and are adjustment factors for each time point, which are used to convert data of different units to a unified scale and adjust the influence of changes in blood glucose and body mass index. The initial value of the adjustment factor is usually set by experience or small-scale test, for example, set to a small random number, and then continuously optimized by the back propagation algorithm in the training process, gradually adjusted to the weight that best reflects the influence of input features on output results. The optimization goal is to make the diabetes risk coefficient output by the neural network as close as possible to the true label in the training data. is the number of continuous data points collected within the time window. is the Sigmoid activation function, defined as: where the input represents the result of the weighted sum in front of the neural network, that is: ; this is a continuous real number, which can be positive or negative. represents a negative exponential transformation of the input, used to construct a smooth and continuous sigmoid function that outputs close to 0 when the input is small and close to 1 when the input is large. The sigmoid function serves to compress this arbitrary real number into the interval (0, 1), so that the output value can be interpreted as a probability or risk level, making it more suitable for representing the diabetes risk factor. The bias term is a trainable parameter that acts as a "baseline level" to adjust the overall function output. It does not depend on the input data, but is randomly set during model initialization and updated and optimized through the backpropagation algorithm together with the adjustment factor during the training process. The purpose of setting the bias term is to help the model fit the training data more flexibly and improve the overall prediction ability, especially when the input data distribution is asymmetric or there is a systematic bias.

[0030] The model receives standardized data of blood glucose values and body mass index and , and assigns weights and to each time point, respectively, reflecting the degree of influence of each variable on the risk. Then, each input variable is multiplied by the corresponding weight and added to the bias term to form a weighted sum value representing the linear combination of diabetes risk. Then, this weighted sum value is input into the Sigmoid activation function, which maps it to between 0 and 1, outputting a risk probability value. This risk probability value represents the probability of developing diabetes, reflecting the individual's health risk at that time point. If the output is close to 1, it indicates high risk; close to 0, it indicates low risk. Through this calculation process, the model can predict the probability of future diabetes based on the individual's blood glucose and body weight changes.

[0031] If patient A's current blood glucose value is 155 mg / dL and body mass index is 29.0 kg / m², the standardized values of historical blood glucose and body mass index data are , , and the adjustment factors are , , and the bias term is , then the neural network calculation process is as follows: ; ; Finally, the diabetes risk factor of patient A is obtained as .

[0032] Please refer to Figure 5The specific steps for comparing the diabetes risk coefficient with the preset risk threshold, evaluating the risk state and potential disease exacerbation of the diabetes patient, and generating diabetes risk warning information are as follows: S401: Compare the diabetes risk coefficient with the preset risk threshold to determine whether the patient is in a risk state, and generate a risk state judgment result based on the comparison result; The calculation of the diabetes risk coefficient is based on known health data such as blood glucose level, body mass index, age, and other factors. These data are combined through certain statistical and algorithmic models to obtain a specific risk coefficient value. By comparing the risk coefficient with the risk threshold (such as a preset value of 0.7 or 0.8), it can be determined whether the patient is in a high-risk state. For example, assume that a patient's diabetes risk coefficient is 0.78 and the preset threshold is 0.8. When the coefficient is less than the threshold, the patient is considered to be in a low-risk state, and vice versa. This judgment process can be performed through specific data combined with logical judgment. For example, if a patient's blood glucose level is 9.2 mmol / L, body mass index is 30 (which falls within the obese range), and age is 50 years old, these data are input into the risk assessment model, and an algorithm is used to obtain a risk coefficient, such as 0.78. If the preset risk threshold is 0.8, the patient's risk state will be judged to be relatively low because 0.78 is lower than the threshold of 0.8. The final result is that the patient's diabetes risk is evaluated as low and is in a relatively low-risk state. This process can be performed by setting up a risk model in an electronic health record system and automatically executing it to assess the risk of each patient in real time and generate a corresponding report.

[0033] S402: Based on the risk state judgment result, evaluate the patient's potential disease exacerbation or complication risk, identify physiological changes that lead to worsening of diabetes symptoms, analyze the patient's risk development trend by comparing the current blood glucose level and body mass index, and generate diabetes risk warning information; According to the judgment result of the diabetes risk state, the patient's condition deterioration or complication risk is further evaluated. This evaluation process usually needs to analyze the patient's blood glucose level and body mass index (BMI) to generate early warning information of diabetes risk. For example, if the patient's blood glucose level has risen to 10.5 mmol / L, while the body mass index remains at 29.5, and the patient has a family history, it may mean that the patient's diabetes has entered a potential deterioration state. At this time, the system may analyze the trend of blood glucose level, for example, if the blood glucose is consistently high and the patient's weight gain trend is not effectively controlled, early warning is needed to prevent further deterioration of the condition. In the evaluation process, first, the patient's current blood glucose level is analyzed to see if it is outside the normal range. Generally, the fasting blood glucose level should be maintained between 4.4 and 6.1 mmol / L, and if it exceeds this range, the patient may be at risk of diabetes or pre-diabetes; similarly, if the body mass index exceeds 25, the patient is considered overweight or obese, which makes them more likely to develop diabetes. By analyzing the trend of blood glucose level and body mass index, the deterioration of the patient's condition can be analyzed to generate early warning information of diabetes, alerting medical personnel or the patient to intervene early and prevent complications. This analysis process can also be automatically performed by an algorithm in an electronic health system, which will judge different risk data intervals and give corresponding warnings or recommendations.

[0034] Please refer to Figure 6 It also includes step S5, based on the diabetes risk early warning information, obtaining the regular physical examination data of the diabetes patient to monitor the change of health status, re-evaluating the risk state of the diabetes patient according to the updated physical examination data, and generating an updated risk assessment result; The risk assessment result includes the updated risk state, the latest health data evaluation result, and the risk level adjustment; Based on the diabetes risk early warning information, the risk state of the diabetes patient is updated in real time, and the specific steps of generating the updated risk assessment result are as follows: S501: Based on the diabetes risk early warning information, obtain the regular physical examination data of the diabetes patient and monitor and record, including the latest blood glucose level and body mass index health indicators, to generate regular physical examination health data; According to the diabetes risk warning information, the collection and monitoring of regular physical examination data generally involve the patient's latest blood glucose level and body mass index and other health indicators. Regular physical examination data is generally entered into the system by the hospital or health management platform during the patient's visit, including blood glucose test results (such as fasting blood glucose value), body mass index (BMI), blood pressure, blood lipids, etc. Blood glucose level is tested by collecting blood samples, usually measured by standard blood glucose testing instruments, with a value of 8.4 mmol / L. The patient's weight and height are used to calculate BMI, if the weight is 85 kg and the height is 1.75 m, the calculation formula is BMI = weight (kg) / height 2 (m 2), so BMI = 85 / (1.75) 2 ≈ 27.8, which belongs to the overweight category. After collecting these physical examination data, the data will be entered into the health management system or electronic health record (EHR), generating a corresponding regular physical examination report. For example, assuming that the latest blood glucose level of a diabetic patient is 8.4 mmol / L and the BMI is 27.8, the patient's information will be recorded by the system and a regular physical examination health data report will be automatically generated, which contains the current blood glucose value, BMI value and other related health information. This report will be retained in the system for subsequent health monitoring and tracking.

[0035] S502: Based on the regular physical examination health data, re-evaluate the risk status of the diabetic patient, compare the updated health indicators with the previous diabetes risk coefficient, judge whether the patient's health status has changed, and adjust the diabetes risk status according to the change, generate an updated risk assessment result; Compare the current collected health indicators with the previous diabetes risk coefficient, for example, by the model set in the early stage (may combine blood glucose, BMI, etc.), re-calculate the risk coefficient. If the updated health indicators show that the patient's blood glucose level has risen, such as from 6.8 mmol / L last time to 8.4 mmol / L, or the BMI value has increased, it means that the risk status may change. At this time, the system will compare the new health data with the previous risk coefficient to analyze whether it exceeds the preset risk threshold. For example, assuming that the preset risk coefficient threshold is 0.8, if the patient's latest calculation result makes the risk coefficient rise from 0.75 to 0.85, the system will identify that the patient's risk status becomes more serious. At this time, the system will automatically adjust the patient's risk assessment result, generate an updated risk assessment report, and record it in the electronic health system, so that the patient and medical staff can see the change of risk status through the report. This evaluation process actually involves the re-calculation of multiple factors in the model, which may involve weighted average or other statistical processing methods of blood glucose, BMI and other health data. Through this evaluation method, the dynamic changes of the patient's health can be grasped in real time, and corresponding health interventions can be made.

[0036] The above merely describes the preferred embodiments of the present application, but does not limit the present application in other forms. Any person skilled in the art can make changes or modifications to the above disclosed technical contents into equivalent embodiments with equivalent changes, and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application without departing from the technical solution content of the present application still falls within the protection scope of the present application.

Claims

1. An intelligent risk prediction method based on neural network, characterized in that: The following steps are involved: S1: Collect blood glucose levels and body mass index (BMI) of diabetic patients, cluster the standardized data using an unsupervised learning algorithm, divide them into multiple risk groups based on blood glucose and BMI, identify differences in health characteristics within each group, and generate diabetes risk analysis results; S2: Collect the historical blood glucose level and body mass index change paths of the diabetic patient group in the specified period before the onset of the disease in the diabetes risk analysis results, calculate the risk transfer index using time series analysis, identify the typical risk transfer path before the onset of the disease, and generate the diabetes transfer trajectory analysis results; S3: Using the diabetes risk analysis results as input to a neural network model for deep learning, and combining the historical blood glucose level and body mass index change paths in the diabetes transfer trajectory analysis results to calculate the corresponding diabetes risk coefficient; S4: Compare the diabetes risk coefficient with a preset risk threshold, evaluate the risk status and potential disease progression of the diabetic patient, and generate diabetes risk warning information.

2. The neural network-based intelligent risk prediction method according to claim 1, characterized in that: The diabetes risk analysis results include blood glucose level, body mass index, healthy group classification, and potential risk patterns; the diabetes transfer trajectory analysis results include risk transfer index, blood glucose fluctuations before onset, body mass index fluctuations before onset, typical risk transfer paths, and physiological changes that trigger diabetes; the diabetes risk coefficient includes current blood glucose level, current body mass index, historical blood glucose level changes, historical body mass index changes, and fluctuation comparison; the diabetes risk warning information includes current risk status, potential risk of disease worsening, comorbidity risk, and health management warning signals.

3. The neural network-based intelligent risk prediction method according to claim 1, characterized in that: The specific steps for collecting blood glucose levels and body mass index (BMI) of diabetic patients and clustering the standardized data using an unsupervised learning algorithm are as follows: S101: Collecting blood glucose level and body mass index data of diabetic patients, standardizing the data, clustering the blood glucose level and body mass index of each diabetic patient based on the standardized data, obtaining the center point, mean, and variance statistics of the clustered data distribution, and generating statistical information of a diabetic healthy group; S102: Analyzing the clustered patient data using an unsupervised learning algorithm based on the statistical information of the healthy diabetes population, dividing the patients into multiple groups based on the distribution of blood glucose levels and body mass index, each group representing a potential diabetes risk analysis result, and generating a diabetes risk group division result; S103: Based on the diabetes risk group division results, the statistical characteristic values ​​of each group are compared with the risk factors for diabetes, the differences in health characteristics exhibited by the patient groups during the diabetes development process are identified, and the diabetes risk analysis results are generated.

4. The neural network-based intelligent risk prediction method according to claim 1, characterized in that: The specific steps for collecting the historical blood glucose level and body mass index change paths of the diabetic patient group in the specified period before the onset of the disease in the diabetes risk analysis results, calculating the risk transfer index using time series analysis, identifying the typical risk transfer path before the onset of the disease, and generating the diabetes transfer trajectory analysis results are as follows: S201: Collecting historical data on blood glucose levels and body mass index of a group of diabetic patients in the diabetes risk analysis results for a specified period before the onset of the disease, arranging the data in time series, and generating historical blood glucose and body mass index change data by arranging and converting the data; S202: Based on the historical blood glucose and body mass index change data, a time series analysis method is used to calculate a risk transfer index, and by comparing the fluctuations of the patient's blood glucose and body mass index over a specified period, key fluctuation characteristics are extracted to obtain a fluctuation characteristic analysis result; S203: Based on the fluctuation characteristic analysis results, identify the typical risk transfer path of each diabetic patient before the onset of the disease, determine the relationship between the fluctuation pattern of health data and diabetes risk, generate diabetes transfer trajectory analysis results, and identify the key physiological changes that cause diabetes.

5. The neural network-based intelligent risk prediction method according to claim 4, characterized in that: For calculating the risk transfer index , using the formula: ； in, and is the adjustment factor, , is the baseline value of blood sugar changes, is the baseline value of body mass index change, represents the total number of time points in the analysis period, and represent the difference between consecutive time points, Indicates a point in time.

6. The neural network-based intelligent risk prediction method according to claim 1, characterized in that: The specific steps for calculating the corresponding diabetes risk coefficient by using the diabetes risk analysis results as input to a neural network model for deep learning and combining the historical blood glucose level and body mass index change paths in the diabetes transfer trajectory analysis results are as follows: S301: Integrate the diabetes risk analysis results and the diabetes transfer trajectory analysis results, extract the current blood glucose level and body mass index data of the diabetic patients, and the historical blood glucose and body mass index change paths of each patient, and pre-process them to generate standardized input data; S302: The standardized input data is input into a neural network for training. The neural network will combine the current blood sugar level, body mass index and historical change path to learn the patient's health pattern and calculate the diabetes risk coefficient of each patient to quantify the diabetes risk of the patient under the patient's current health status.

7. The neural network-based intelligent risk prediction method according to claim 6, characterized in that: To calculate the diabetes risk factor for each patient , using the formula: ； in, Indicates the Blood sugar at time point, Indicates the Body mass index at time point, and is the adjustment factor for each time point, is the Sigmoid activation function, is the bias term, is the number of consecutive data points collected within the time window.

8. The neural network-based intelligent risk prediction method according to claim 1, characterized in that: The specific steps of comparing the diabetes risk coefficient with a preset risk threshold, evaluating the risk status and potential disease progression of the diabetic patient, and generating diabetes risk warning information are as follows: S401: Comparing the diabetes risk coefficient with a preset risk threshold to determine whether the patient is in a risk state, and generating a risk state judgment result based on the comparison result; S402: Based on the risk status judgment result, the patient's potential risk of disease worsening or complications is assessed, the physiological changes that lead to worsening diabetes symptoms are identified, and the patient's risk development trend is analyzed by comparing the current blood sugar level and body mass index to generate diabetes risk warning information.

9. The neural network-based intelligent risk prediction method according to claim 1, characterized in that: The method further includes step S5 of updating the risk status of the diabetic patient in real time based on the diabetes risk warning information to generate an updated risk assessment result; The risk assessment results include updated risk status, latest health data assessment results, and risk level adjustments.

10. The neural network-based intelligent risk prediction method according to claim 1, characterized in that: Based on the diabetes risk warning information, the specific steps of updating the risk status of the diabetic patient in real time and generating an updated risk assessment result are as follows: S501: Based on the diabetes risk warning information, obtain regular physical examination data of the diabetic patient and monitor and record it, including the latest health indicators of blood sugar level and body mass index, to generate regular physical examination health data; S502: Based on the regular physical examination health data, re-evaluate the risk status of the diabetic patient, compare the updated health indicators with the previous diabetes risk coefficient, determine whether the patient's health status has changed, and adjust the diabetes risk status according to the changes to generate an updated risk assessment result.

Images