A chronic disease risk prediction method and system based on multi-modal data fusion

Abstract

This invention discloses a method and system for predicting chronic disease risk based on multimodal data fusion, relating to the field of chronic disease prediction. It includes: an upload module for uploading patient TCM syndrome description data and Western medicine physiochemical test data, classifying and storing them according to data type, and simultaneously establishing multi-source data index identifiers; and a processing module for performing semantic standardization processing on the classified TCM syndrome data and outlier filtering on the Western medicine physiochemical indicator data to output normalized multimodal data. This invention integrates TCM syndrome and Western medicine physiochemical test data, standardizes semantic expressions, accurately filters out abnormal data, fully explores the correlation features and time-series change patterns between the two, dynamically allocates feature weights to achieve deep fusion, and constructs a TCM-Western medicine collaborative prediction logic, making the quantification of chronic disease risk levels more reasonable.

Description

Technical Field

[0001] This invention relates to the field of chronic disease prediction technology, specifically to a method and system for predicting chronic disease risk based on multimodal data fusion. Background Technology

[0002] Chronic disease risk prediction integrates personal health data, lifestyle habits, and environmental factors, and uses machine learning algorithms to build predictive models to quickly assess disease risk, generate personalized early warning information, and provide technical support for early disease intervention and health management.

[0003] The invention patent application with application number 202510612722.0 discloses an AI-based chronic disease biochemical data analysis system. This application aims to solve the problem that "some studies attempt to extract chronic disease-related symptoms and treatment information from electronic health records based on natural language processing technology, and use neural network models for classification and display. However, such methods are mostly limited to text mining and static knowledge extraction, lack the ability to jointly model patients' behavioral habits and physiological indicators, and have failed to achieve the time series expression of behavioral risk scores and the adaptive update mechanism of model fusion. Moreover, the existing technology has not achieved deep integration of biochemical test data and natural language information, lacks a multimodal data evaluation mechanism, and its risk assessment model has limited comprehensiveness and accuracy."

[0004] However, in existing technologies, most chronic disease risk predictions rely on quantitative data from Western medical examinations as the main basis for continuous monitoring and prediction, without incorporating traditional Chinese medicine's syndrome data for chronic diseases. This results in a single basis for chronic disease prediction, making it difficult to effectively predict patients' chronic diseases.

[0005] To this end, we propose a method and system for predicting chronic disease risk based on multimodal data fusion. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a method and system for predicting chronic disease risk based on multimodal data fusion, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a chronic disease risk prediction system based on multimodal data fusion, comprising: The system comprises the following modules: Upload module for uploading patient TCM syndrome descriptions and Western medicine physiochemical test data, classifying and storing them according to data type, and simultaneously establishing multi-source data indexes; Processing module for semantic standardization of the classified TCM syndrome data and outlier filtering of the Western medicine physiochemical indicator data to output normalized multimodal data; Extraction module for extracting syndrome-physiochemical correlation features from the normalized data and constructing a multi-dimensional feature set; Fusion module for setting dynamic allocation logic for TCM and Western medicine feature weights, performing hierarchical fusion of the multi-dimensional feature set, and generating a unified feature vector; Prediction module for constructing a TCM-Western medicine collaborative prediction logic based on the unified feature vector, and performing quantitative extrapolation of chronic disease risk levels based on the prediction logic; and Feedback module for receiving the quantitative extrapolation results from the prediction module and pushing the results to a preset receiving end. The upload module is interconnected with the processing module and the extraction module via a wireless network. The extraction module is interconnected with the fusion module via a wireless network. The fusion module is interconnected with the prediction module and the feedback module via a wireless network.

[0008] Furthermore, in the data classification and storage stage of the upload module, independent storage directories are established according to the representation type of TCM syndrome data and the project attributes of Western medicine physiochemical test data, and the data are sorted in chronological order according to the data collection timestamp within the directories; The multi-source data index identifier consists of a unique patient identification code, a data type identifier, a data acquisition device number, a timestamp field, and a check bit. The data type identifier distinguishes between traditional Chinese medicine syndrome data and Western medicine physicochemical test data through binary encoding, and the check bit is used to verify the integrity of the index encoding.

[0009] Furthermore, the semantic standardization process in the processing module is as follows: Semantic segmentation was performed on TCM syndrome description data to extract core symptom terms; Based on a pre-set dictionary of TCM syndrome synonyms, synonyms or near-synonyms are uniformly mapped to standard terms; Referring to the TCM syndrome classification standards, complete the category attribution labeling for standard terms; In the outlier filtering stage of the processing module, the filtering operation is performed by calculating the outlier determination coefficient of the data. ; In the formula: This is the anomaly detection coefficient; , These are the statistical characteristic weights and the clinical rationality weights, respectively. For individual indicator data from Western medical physicochemical tests; This is the average of the data for this indicator uploaded in the same batch; The standard deviation of the data in the same batch; This is the corresponding clinical standard reference value for this indicator; This represents the clinically permissible range of fluctuation for this indicator; when If the value exceeds the preset abnormal threshold, the data is determined to be an abnormal value and marked for removal.

[0010] Furthermore, in the extraction module, when extracting the syndrome-physicochemical correlation features, the core syndrome factors are first extracted from the normalized TCM syndrome data, and key physicochemical indicators are extracted from the Western medicine physicochemical data. Calculate the correlation strength coefficient between each syndrome factor and each physicochemical indicator, and select syndrome-physicochemical combinations with correlation strength exceeding a preset threshold as correlation features; The multi-dimensional feature set specifically includes a subset of syndrome features, a subset of physicochemical features, a subset of syndrome-physicochemical correlation features, and a subset of temporal change features. The temporal change feature subset is constructed by extracting the changing trend and fluctuation amplitude of the same indicator in different collection periods.

[0011] Furthermore, the dynamic allocation logic for the weights of traditional Chinese medicine and Western medicine features in the fusion module is as follows: ; In the formula: The dynamic weight of the i-th feature; , , These are the weighting coefficients for discrimination, stability, and clinical relevance, respectively. , , The sum is 1; This is the average of the discrimination scores for all features; This is the average value of the stability of all features; This represents the average clinical relevance of all features; The total number of features participating in weight allocation; Let be the discriminative power of the i-th feature; The stability of the i-th feature; For the clinical relevance of the i-th feature, Similarly; The hierarchical fusion is performed sequentially as follows: feature layer, feature set layer, and semantic layer. The feature layer standardizes and normalizes individual features; the feature set layer performs weighted fusion of Chinese and Western medicine feature subsets based on dynamic weights to obtain Chinese and Western medicine feature set vectors; the semantic layer transforms the Chinese and Western medicine feature set vectors into feature vectors under a unified semantic space through a feature semantic mapping matrix to complete the fusion.

[0012] Furthermore, the prediction module's integrated traditional Chinese and Western medicine prediction logic is as follows: ; In the formula: Quantification of chronic disease risk levels; , , The integrated weights are based on fundamental risk, dialectical correction, and time series optimization. This represents the basic risk value for Western medicine. This is a correction value for TCM syndrome differentiation; This refers to the rate of change of time-series risk. The larger the value, the higher the risk; conversely, the smaller the value, the lower the risk. System users continuously monitor this data. The changing trend or presupposition and Risk prediction is conducted in this way.

[0013] Furthermore, the aforementioned , The calculation formula is: ; In the formula: This represents the total number of physicochemical indicators in Western medicine. Let be the weight of the j-th Western medicine physicochemical indicator; For indicator risk mapping function; This represents the total number of TCM syndrome characteristics; The weight of the k-th TCM syndrome feature; For syndrome differentiation risk function; in, >0, and ; >0, and .

[0014] Furthermore, the expressions for the indicator risk mapping function and the syndrome differentiation risk function are as follows: ; In the formula: The risk amplification coefficient for the j-th indicator; The data is the normalized data for the j-th Western medicine physicochemical indicator; This represents the clinical standard reference value corresponding to the j-th indicator; The clinically permissible fluctuation range for the j-th indicator; Let be the nonlinear exponent of the j-th index; Let be the risk threshold parameter for the j-th indicator; This is the direction correction coefficient for the j-th index; This is a sign function; it returns 1 when the value inside the parentheses is positive, -1 when it is negative, and 0 when it is zero. The basic risk coefficient for the kth syndrome; is the standardized intensity value of the k-th TCM syndrome feature; Let be the intensity saturation coefficient of the k-th syndrome; Let be the intensity sensitivity coefficient of the k-th syndrome; This represents the number of other syndromes that have a synergistic effect with the k-th syndrome. Let be the co-risk coefficient between the k-th syndrome and the l-th co-syndrome; is the standardized intensity value of the l-th co-symptom.

[0015] On the other hand, a chronic disease risk prediction method based on multimodal data fusion includes: Upload patient TCM syndrome description data and Western medicine physiochemical test data, store them according to data type, and establish a multi-source data index identifier containing fields such as patient unique identification code and data type identifier; perform semantic word segmentation, synonym mapping, and category labeling standardization processing on TCM syndrome data, filter outliers in Western medicine physiochemical data through anomaly judgment coefficient, and output normalized multimodal data; extract core TCM syndrome factors and key Western medicine physiochemical indicators, calculate correlation strength coefficient to filter effective correlation features, and construct a multi-dimensional feature set based on time-series change features; dynamically allocate TCM and Western medicine feature weights based on feature discrimination, stability, and clinical relevance, and complete hierarchical fusion through feature layer normalization, feature set layer weighting, and semantic layer mapping to generate a unified feature vector; set up TCM and Western medicine collaborative prediction logic, integrate Western medicine basic risk value, TCM syndrome correction value, and time-series risk change rate to quantify and infer chronic disease risk level; receive the quantification results of chronic disease risk level and push the quantification results to the designated receiving end.

[0016] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention integrates TCM syndromes and Western medicine physicochemical test data, standardizes semantic expressions, accurately screens abnormal data, fully explores the correlation characteristics and time-series changes of the two, dynamically allocates feature weights to achieve deep integration, and constructs a TCM-Western medicine collaborative prediction logic. This makes the quantification of chronic disease risk levels more reasonable, taking into account both the reliability of data statistics and actual clinical needs, while also capturing the dynamic changing trends of indicators, improving the accuracy and pertinence of risk prediction, providing a comprehensive reference for early warning and intervention of chronic diseases, reducing missed and misjudgments, and improving the efficiency and accuracy of health management. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 This is a schematic diagram of the structure of a chronic disease risk prediction system based on multimodal data fusion; Figure 2 This is a flowchart illustrating a chronic disease risk prediction method based on multimodal data fusion. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] The present invention will be further described below with reference to embodiments.

[0021] Example 1: This embodiment presents a chronic disease risk prediction system based on multimodal data fusion, such as... Figure 1 As shown, it includes: The upload module is used to upload patients' TCM syndrome description data and Western medicine physicochemical test data, classify and store them according to data type, and simultaneously establish multi-source data index identification; In the data classification and storage stage of the upload module, separate storage directories are established according to the representation type of TCM syndrome data and the project attributes of Western medicine physiochemical test data. The data in the directories are sorted in time sequence according to the data collection timestamp. The multi-source data index identifier consists of a unique patient identification code, a data type identifier bit, an acquisition device number, a timestamp field, and a check bit. The data type identifier bit uses binary encoding to distinguish between traditional Chinese medicine syndrome data and Western medicine physicochemical test data, and the check bit is used to verify the integrity of the index code. Among them, the types of manifestations include symptom descriptions, physical signs records, and diagnostic conclusions, and the project attributes include blood tests, imaging tests, and functional monitoring; The processing module is used to perform semantic standardization on the TCM syndrome data stored in categories and to filter out outliers on the Western medicine physicochemical index data in order to output normalized multimodal data. The semantic standardization process in the processing module is as follows: Semantic segmentation was performed on TCM syndrome description data to extract core symptom terms; Based on a pre-set dictionary of TCM syndrome synonyms, synonyms or near-synonyms are uniformly mapped to standard terms; Referring to the TCM syndrome classification standards, complete the category attribution labeling for standard terms; In the outlier filtering phase of the processing module, the filtering operation is performed by calculating the outlier determination coefficient of the data. ; In the formula: This is the anomaly detection coefficient; , These are the statistical characteristic weights and the clinical rationality weights, respectively. For individual indicator data from Western medical physicochemical tests; This is the average of the data for this indicator uploaded in the same batch; The standard deviation of the data in the same batch; This is the corresponding clinical standard reference value for this indicator; This represents the clinically permissible range of fluctuation for this indicator; The above formula combines the statistical characteristics of data with the clinical rationality of clinical data to identify outliers. By allocating the weights of statistical characteristics and clinical rationality, it not only refers to the statistical information such as the mean and standard deviation of the same batch of data, but also associates the corresponding clinical standard reference values ​​and allowable fluctuation ranges of the indicators. At the same time, it dynamically adjusts the weight ratio of the two based on the data sample size, distribution uniformity, and the clarity of the clinical reference standards and the width of the allowable fluctuation range. This effectively avoids misjudgment caused by relying solely on statistical or clinical dimensions, so as to achieve accurate screening of outliers under different data quality and clinical scenarios. when If the value exceeds the preset abnormal threshold, the data is determined to be an abnormal value and marked for removal. in, , The sum is 1, and , The value range is preset to [0.1, 0.9]. When the sample size of the same batch of Western medicine physicochemical test data is large, the data distribution is uniform, and the statistical results are highly reliable, the range is suitable for testing. The larger the value, the better when the sample size of the same batch of data is small, the distribution is discrete, and the reliability of the statistical results is low. The smaller the value, the better, especially when the clinical reference standard for the test indicator is clear, the acceptable range of clinical fluctuation is narrow, and the clinical dimension has a more critical impact on the judgment of abnormalities. The larger the value, the better when the clinical reference standard for the test indicator is vague, the allowable range of fluctuation is wide, and the impact of clinical dimensions is small. The smaller the value; The extraction module is used to extract the symptom-physicochemical correlation features from the normalized data and construct a multi-dimensional feature set; In the extraction module, the syndrome-physicochemical correlation features are extracted by first extracting the core syndrome factors from the normalized TCM syndrome data and then extracting the key physicochemical indicators from the Western medicine physicochemical data. Calculate the correlation strength coefficient between each syndrome factor and each physicochemical indicator, and select syndrome-physicochemical combinations with correlation strength exceeding a preset threshold as correlation features; The multi-dimensional feature set specifically includes a subset of syndrome features, a subset of physicochemical features, a subset of syndrome-physicochemical correlation features, and a subset of temporal change features. The subset of temporal change features is constructed by extracting the changing trends and fluctuation amplitudes of the same indicator in different collection periods. Among them, the core syndrome factors include the main symptom characteristics, secondary symptom characteristics and syndrome combination characteristics, and the key physicochemical indicators include core detection indicators, derived indicators and indicator change rate characteristics. The core syndrome factors were obtained by statistically analyzing the frequency of occurrence of each syndrome term in the TCM syndrome data, combined with the pre-set clinical syndrome weights and the degree of correlation between syndromes. The key physicochemical indicators were determined by evaluating the clinical diagnostic contribution, data discrimination, and cross-period variation stability of each indicator in the Western medicine physicochemical data. Correlation strength coefficient between syndrome factors and physicochemical indicators ; In the formula: These are statistical correlation weights, feature attribute weights, and temporal collaboration weights, respectively. Let be the mutual information value between the i-th syndrome factor and the j-th physicochemical index; The average value of the mutual information values ​​of all syndrome factors-physicochemical index combinations; Let be the intensity value of the i-th syndrome factor; Let be the deviation of the j-th physicochemical index; Let be the temporal synergy coefficient between the i-th syndrome factor and the j-th physicochemical index; It is the average value of the temporal synergistic coefficients of all syndrome factors-physicochemical index combinations; The above formula comprehensively considers three core dimensions: statistical correlation between syndrome factors and physicochemical indicators, feature attributes and temporal synergistic effects. Through the differentiated allocation of statistical correlation weight, feature attribute weight, and temporal synergistic weight, it incorporates key information such as mutual information value, syndrome factor intensity value, deviation of physicochemical indicators and temporal synergistic coefficient. At the same time, it normalizes the formula with the mean of all combinations, which can accurately capture the multivariate correlation between syndrome and physicochemical indicators in different scenarios, making the correlation feature selection more in line with clinical practice and data change patterns. in, ∈[0,1], when the statistical correlation (characterized by mutual information value) between the syndrome factor and the physicochemical index has a higher priority in determining the correlation strength, the value is larger; when the influence of this statistical correlation does not need to be considered, the value is smaller. ∈[0,1], the value is larger when the strength attribute of the syndrome factor and the deviation attribute of the physicochemical index are more critical to the determination of the correlation strength, and the value is smaller when the correlation contribution of the feature's own attribute needs to be weakened. The value of ∈[0,1] is larger when the consistency of the changing trends of the syndrome factors and physicochemical indicators within the continuous collection period plays a more prominent role in determining the correlation strength, and smaller when the influence of the consistency of time sequence changes does not need to be emphasized. The higher the frequency of occurrence and clinical severity of the syndrome factor, the larger the value; conversely, the lower the frequency and clinical severity, the smaller the value. ∈[0,1], is obtained by calculating the absolute difference between the actual detected value and the clinical reference value of the physicochemical indicator, dividing it by the clinically permissible fluctuation range of the indicator, and then normalizing it; ∈[0,1], calculated based on the consistency of the changing trends of the two within the continuous acquisition period; The fusion module is used to set the dynamic allocation logic of the weights of Chinese and Western medicine features, and to perform hierarchical fusion of multi-dimensional feature sets to generate a unified feature vector. The dynamic allocation logic for the weights of traditional Chinese medicine and Western medicine features in the fusion module is as follows: ; In the formula: The dynamic weight of the i-th feature; , , These are the weighting coefficients for discrimination, stability, and clinical relevance, respectively. , , The sum is 1; This is the average of the discrimination scores for all features; This is the average value of the stability of all features; This represents the average clinical relevance of all features; The total number of features participating in weight allocation; Let be the discriminative power of the i-th feature; The stability of the i-th feature; For the clinical relevance of the i-th feature, Similarly; The above formula is based on three core attributes of feature prediction of chronic disease risk: discrimination, stability and clinical relevance. It normalizes the features by introducing the mean of each attribute and combines the weight coefficients of discrimination, stability and clinical relevance with the comprehensive calculation results of all features involved in the weight allocation to dynamically adapt the feature weights. This fully considers the unique value of each feature and can reasonably allocate weights according to the actual contribution of different features, thereby avoiding the imbalance of feature importance caused by equal weights. Hierarchical fusion is performed sequentially as follows: feature layer, feature set layer, and semantic layer. The feature layer standardizes and normalizes individual features; the feature set layer performs weighted fusion of the TCM and Western medicine feature subsets based on dynamic weights to obtain the TCM and Western medicine feature set vectors; the semantic layer transforms the TCM and Western medicine feature set vectors into feature vectors under a unified semantic space through a feature semantic mapping matrix to complete the fusion. in, The degree of correlation between features and chronic disease risk levels is determined by calculating the mutual information algorithm; The characteristic variation coefficients are expressed inversely over three consecutive acquisition cycles; Based on the predetermined clinical feature importance determination, within the predetermined range, the higher the clinical feature importance of the i-th feature, the larger the value, and vice versa; The prediction module is used to construct a prediction logic that integrates traditional Chinese and Western medicine based on a unified feature vector, and to perform quantitative deduction of chronic disease risk levels based on the prediction logic. The prediction logic of the integrated traditional Chinese and Western medicine approach in the prediction module is as follows: ; In the formula: Quantification of chronic disease risk levels; , , The integrated weights are based on fundamental risk, dialectical correction, and time series optimization. This represents the basic risk value for Western medicine. This is a correction value for TCM syndrome differentiation; This refers to the rate of change of time-series risk. The above formula integrates three key dimensions: basic risk assessment by Western medicine, syndrome differentiation and correction by traditional Chinese medicine, and time-series risk change trend. It dynamically adjusts the weights of basic risk, syndrome differentiation and correction, and time-series optimization. At the same time, it obtains the time-series risk change rate by fitting the slope of the risk values ​​of the most recent three periods. It distinguishes the limitations of isolated prediction by traditional Chinese and Western medicine and the neglect of time-series changes, and completes a comprehensive and dynamic quantitative assessment of the risk level of chronic diseases. The larger the value, the higher the risk; conversely, the smaller the value, the lower the risk. System users continuously monitor this data. The changing trend or presupposition and Risk prediction is conducted in a manner that allows for risk assessment. in, , , The values ​​of are all in the range of (0,1), and The value is positively correlated with the completeness, reliability, and clinical reference value of the Western medicine physiochemical indicator data. That is, when the Western medicine physiochemical indicator data is highly complete, the test results are reliable, and the clinical correlation with the target chronic disease is strong, the value is positively correlated. The larger the value, the smaller the value; The magnitude of the value is positively correlated with the standardization degree of TCM syndrome data, the clarity of syndrome differentiation, and the strength of the syndrome-disease correspondence. That is, when the TCM syndrome description is standardized, the syndrome differentiation conclusion is clear, and the correspondence between the core syndrome and the target chronic disease is significant, the value is positively correlated with the disease. The larger the value, the smaller the value; The value is positively correlated with the continuity and significant fluctuation trend of risk-related data in the most recent three collection periods, i.e., when the time series data is complete and unbroken, and the changing trends of risk-related indicators are clear and traceable. The larger the value, the smaller the value; The slope was calculated by fitting the risk values ​​from the most recent three data collection periods. , The calculation formula is: ; In the formula: This represents the total number of physicochemical indicators in Western medicine. Let be the weight of the j-th Western medicine physicochemical indicator; For indicator risk mapping function; This represents the total number of TCM syndrome characteristics; The weight of the k-th TCM syndrome feature; For syndrome differentiation risk function; In the above formula, for the basic risk value of Western medicine, each physicochemical indicator is assigned a weight that matches its contribution to clinical diagnosis, and the risk of each indicator is weighted and aggregated by combining the indicator risk mapping function; for the TCM syndrome differentiation correction value, a specific weight is assigned according to the clinical importance of each syndrome feature, and the comprehensive calculation of the risk of each syndrome is completed by combining the syndrome differentiation risk function, so as to accurately quantify the contribution of the core risk factors of both Chinese and Western medicine, ensure the scientificity and pertinence of the single-dimensional risk assessment of Chinese and Western medicine, and lay the foundation for subsequent collaborative risk prediction of Chinese and Western medicine. in, >0, and ; >0, and ; The expressions for the indicator risk mapping function and the syndrome differentiation risk function are as follows: ; In the formula: The risk amplification coefficient for the j-th indicator; The data is the normalized data for the j-th Western medicine physicochemical indicator; This represents the clinical standard reference value corresponding to the j-th indicator; The clinically permissible fluctuation range for the j-th indicator; Let be the nonlinear exponent of the j-th index; Let be the risk threshold parameter for the j-th indicator; This is the direction correction coefficient for the j-th index; This is a sign function; it returns 1 when the value inside the parentheses is positive, -1 when it is negative, and 0 when it is zero. In the above formula, for the risk mapping of Western medicine physicochemical indicators, a risk amplification coefficient is set in combination with the importance of the indicator to chronic diseases, a nonlinear index is determined based on the risk change trend after deviating from the reference value, a direction correction coefficient is set according to the correlation between the indicator and the risk, and the difference between the actual value and the reference value of the indicator, the allowable fluctuation range and the risk threshold parameters are incorporated to complete the accurate and nonlinear transformation of indicator risk. For the risk function of TCM syndrome differentiation, the basic risk coefficient is set according to the primary and secondary status of the syndrome in the differentiation. The relevant parameters are determined by combining the intensity of clinical manifestations, risk saturation characteristics, and importance of influence. At the same time, the synergistic effect and synergistic strength with other syndromes are considered to construct a quantitative model that fits the cumulative law of syndrome risk, so that the single-factor risk mapping of TCM and Western medicine are consistent with their respective medical logic and clinical practice. The basic risk coefficient for the kth syndrome; is the standardized intensity value of the k-th TCM syndrome feature; Let be the intensity saturation coefficient of the k-th syndrome; Let be the intensity sensitivity coefficient of the k-th syndrome; This represents the number of other syndromes that have a synergistic effect with the k-th syndrome. Let be the co-risk coefficient between the k-th syndrome and the l-th co-syndrome; The standardized intensity value of the l-th co-symptom; in, The preset value range is [0.8, 3]. The more critical the Western medicine physicochemical indicators (such as blood pressure and fasting blood glucose) are to the onset or progression of chronic diseases, the larger the value is; the less critical the indicators (such as serum ferritin and other auxiliary indicators) are to the chronic diseases, the smaller the value is. The preset value range is [1, 2.5]. For indicators whose risk needs to increase rapidly after deviating from the clinical reference value (such as uric acid and serum creatinine), the larger the value, the smaller the value for indicators whose risk needs to increase slowly after deviating from the reference value (such as stability indicators such as platelet count in blood routine). The value can only be one of three discrete values: -1, 0, and 1. The higher the value of the indicator, the more positively correlated it is with the risk of chronic diseases. The lower the value of the indicator, the more positively correlated it is with the risk of chronic diseases. The value is 0 when both too high and too low an indicator will increase the risk of chronic diseases. ∈[0.1,1], the larger the value is when the syndrome is the main symptom in the diagnosis of chronic diseases, and the smaller the value is when it is a concurrent symptom; ∈ (0,1), the value is larger when the clinical manifestation of the syndrome is more significant (such as severe syndrome), and smaller when the manifestation is milder (such as mild syndrome); ∈ (0, 1), the larger the value is when the syndrome needs to reach a high intensity to significantly increase the risk of chronic diseases, and the smaller the value is when the mild intensity, i.e., the risk increases rapidly; ≥1 indicates that the more critical the impact of the syndrome on the risk of chronic diseases, the larger the value; and the less critical the impact on the risk, the smaller the value. (0,0.3], the value is larger when the additive effect of the two syndrome combinations on the risk of chronic diseases is more obvious, and the value is smaller when the additive effect of the combination is weaker; The feedback module is used to receive the quantization results from the prediction module and push the quantization results to the preset receiving end. The upload module interacts with the processing module and the extraction module via a wireless network. The extraction module interacts with the fusion module via a wireless network. The fusion module interacts with the prediction module and the feedback module via a wireless network.

[0022] In this embodiment, the upload module uploads the patient's TCM syndrome description data and Western medicine physiochemical test data, classifies and stores them according to data type, and simultaneously establishes multi-source data index identifiers. The processing module runs afterward to perform semantic standardization processing on the classified and stored TCM syndrome data and to filter outout values ​​in the Western medicine physiochemical indicator data to output normalized multimodal data. The extraction module further extracts syndrome-physiochemical correlation features from the normalized data and constructs a multi-dimensional feature set. Then, the fusion module sets the dynamic allocation logic of TCM and Western medicine feature weights and performs hierarchical fusion on the multi-dimensional feature set to generate a unified feature vector. The prediction module constructs a TCM and Western medicine collaborative prediction logic based on the unified feature vector, performs quantitative inference of chronic disease risk level based on the prediction logic, and finally receives the quantitative inference results from the prediction module through the feedback module and pushes the quantitative inference results to the preset receiving end.

[0023] In the above embodiments, the system integrates relevant data from both traditional Chinese and Western medicine to quantify and extrapolate the risk level of chronic diseases. This approach not only improves the accuracy and comprehensiveness of risk prediction but also clearly presents the trend of risk changes, providing a scientific basis for the prevention and control of chronic diseases and thus effectively meeting the actual needs of clinical practice and health management.

[0024] Application example: The XX Community Health Service Center used this system to conduct a type 2 diabetes risk screening for a 45-year-old male resident.

[0025] First, the system uploads data on the resident's Traditional Chinese Medicine (TCM) syndrome descriptions and Western medicine physiochemical test data. TCM syndrome descriptions include symptom records and diagnostic conclusions such as "dry mouth and excessive thirst, increased appetite and hunger, fatigue, and weakness in the lower back and knees." Western medicine physiochemical tests cover blood test data such as fasting blood glucose, glycated hemoglobin, and four lipid profiles. The system also records the data acquisition device number and test time. The system establishes independent storage directories for each category: TCM syndrome description, diagnostic conclusion type, and Western medicine physiochemical test blood test items. These directories are sorted by test timestamps, and a multi-source data index identifier is generated, containing the resident's unique identification code, data type identifier, and device number.

[0026] After entering the processing module, the system performs semantic standardization processing on the TCM syndrome data, mapping synonyms such as "dry mouth and excessive thirst" and "thirst and fondness for drinking" to the standard term "dry mouth" and completing the category assignment labeling; for the Western medicine physicochemical data, outlier screening is performed. After calculation, the anomaly judgment coefficient of one fasting blood glucose data is 0.85, which exceeds the preset threshold of 0.7, and it is judged as an outlier and marked for removal. All other data meet the requirements.

[0027] The extraction module then operates, selecting core syndrome factors from the normalized TCM data, with "dry mouth and excessive hunger" as the primary symptom and "fatigue" as the secondary symptom. It extracts fasting blood glucose and glycated hemoglobin as key physicochemical indicators from the Western medicine data. It calculates the correlation strength coefficient between each syndrome factor and the physicochemical indicators, and selects three syndrome-physicochemical combinations with a correlation strength exceeding 0.7 as correlation features. At the same time, it combines the cross-cycle change trend of the indicators to construct a multi-dimensional feature set that includes syndrome, physicochemical, correlation, and temporal change features.

[0028] The fusion module calculates the weights of each feature according to the dynamic allocation logic. Because the resident's Western medicine test data is complete and has high clinical reference value, the Western medicine feature discrimination and clinical relevance weights are relatively high. After dynamic allocation, the weight of the Western medicine physicochemical feature is 0.62 and the weight of the TCM syndrome feature is 0.38. After feature layer standardization, feature set layer weighted fusion and semantic layer mapping transformation, a unified feature vector is generated.

[0029] The prediction module is based on the prediction logic of integrated traditional Chinese and Western medicine. It combines the basic risk value of Western medicine, the syndrome differentiation correction value of traditional Chinese medicine and the time series risk change rate to calculate the final quantitative value of the resident's type 2 diabetes risk level of 0.73, which belongs to the medium-high risk.

[0030] The feedback module then pushes this prediction result to the community doctor's workstation and the resident's personal health account, prompting the doctor to pay close attention to the resident's blood glucose monitoring, suggesting adjustments to the diet and exercise plan, and reminding the resident to have relevant indicators checked regularly.

[0031] Example 2: At the implementation level, based on Example 1, this example refers to... Figure 2 A further detailed description of the chronic disease risk prediction system based on multimodal data fusion in Example 1 is provided below: A method for predicting chronic disease risk based on multimodal data fusion, comprising: Upload patient TCM syndrome description data and Western medicine physiochemical test data, store them according to data type, and establish a multi-source data index identifier containing fields such as patient unique identification code and data type identifier bit; Standardization processing, including semantic word segmentation, synonym mapping, and category labeling, is performed on TCM syndrome data. Outliers in Western medicine physicochemical data are filtered out using anomaly detection coefficients, and normalized multimodal data is output. Extract core syndrome factors from traditional Chinese medicine and key physicochemical indicators from Western medicine, calculate the correlation strength coefficient to screen effective correlation features, and construct a multi-dimensional feature set based on time-series change features; Based on feature discriminativeness, stability and clinical relevance, the weights of Chinese and Western medicine features are dynamically allocated. Hierarchical fusion is completed through feature layer normalization, feature set layer weighting and semantic layer mapping to generate a unified feature vector. The system sets up a predictive logic that integrates traditional Chinese and Western medicine, combining the basic risk value of Western medicine, the correction value of traditional Chinese medicine syndrome differentiation, and the time series risk change rate to quantify and extrapolate the risk level of chronic diseases. Receive the quantitative results of chronic disease risk levels and push the quantitative results to the designated receiving end.

[0032] In summary, the systems and methods described above integrate traditional Chinese medicine (TCM) syndromes with Western medicine physicochemical test data. By standardizing semantic expressions and accurately screening abnormal data, they fully explore the correlation characteristics and temporal change patterns between the two, dynamically allocate feature weights to achieve deep integration, and construct a TCM-Western medicine collaborative prediction logic. This makes the quantification of chronic disease risk levels more reasonable, taking into account both the reliability of data statistics and actual clinical needs, while also capturing the dynamic change trends of indicators. This improves the accuracy and pertinence of risk prediction, provides a comprehensive reference for early warning and intervention of chronic diseases, reduces missed and misjudgments, and improves the efficiency and accuracy of health management.

[0033] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A chronic disease risk prediction system based on multimodal data fusion, characterized in that, include: The upload module is used to upload patients' TCM syndrome description data and Western medicine physicochemical test data, classify and store them according to data type, and simultaneously establish multi-source data index identification; The processing module is used to perform semantic standardization on the TCM syndrome data stored in categories and to filter out outliers on the Western medicine physicochemical index data in order to output normalized multimodal data. The extraction module is used to extract the symptom-physicochemical correlation features from the normalized data and construct a multi-dimensional feature set; The fusion module is used to set the dynamic allocation logic of the weights of Chinese and Western medicine features, and to perform hierarchical fusion of multi-dimensional feature sets to generate a unified feature vector. The prediction module is used to construct a prediction logic that integrates traditional Chinese and Western medicine based on a unified feature vector, and to perform quantitative inference of chronic disease risk levels based on the prediction logic. The feedback module is used to receive the quantization results from the prediction module and push the quantization results to the preset receiving end.

2. The chronic disease risk prediction system based on multimodal data fusion according to claim 1, characterized in that, In the data classification and storage stage of the upload module, independent storage directories are established according to the representation type of TCM syndrome data and the project attributes of Western medicine physiochemical test data, and the data in the directories are sorted in chronological order according to the data collection timestamp. The multi-source data index identifier consists of a unique patient identification code, a data type identifier, a data acquisition device number, a timestamp field, and a check bit. The data type identifier distinguishes between traditional Chinese medicine syndrome data and Western medicine physicochemical test data through binary encoding, and the check bit is used to verify the integrity of the index encoding.

3. The chronic disease risk prediction system based on multimodal data fusion according to claim 1, characterized in that, The semantic standardization process in the processing module is as follows: Semantic segmentation was performed on TCM syndrome description data to extract core symptom terms; Based on a pre-set dictionary of TCM syndrome synonyms, synonyms or near-synonyms are uniformly mapped to standard terms; Referring to the TCM syndrome classification standards, complete the category attribution labeling for standard terms; In the outlier filtering stage of the processing module, the filtering operation is performed by calculating the outlier determination coefficient of the data. ； In the formula: This is the anomaly detection coefficient; , These are the statistical characteristic weights and the clinical rationality weights, respectively. For individual indicator data from Western medical physicochemical tests; This is the average of the data for this indicator uploaded in the same batch; The standard deviation of the data in the same batch; This is the corresponding clinical standard reference value for this indicator; This represents the clinically permissible range of fluctuation for this indicator; when If the value exceeds the preset abnormal threshold, the data is determined to be an abnormal value and marked for removal.

4. The chronic disease risk prediction system based on multimodal data fusion according to claim 1, characterized in that, In the extraction module, the syndrome-physicochemical correlation features are extracted by first extracting the core syndrome factors from the normalized TCM syndrome data and then extracting the key physicochemical indicators from the Western medicine physicochemical data. Calculate the correlation strength coefficient between each syndrome factor and each physicochemical indicator, and select syndrome-physicochemical combinations with correlation strength exceeding a preset threshold as correlation features; The multi-dimensional feature set specifically includes a subset of syndrome features, a subset of physicochemical features, a subset of syndrome-physicochemical correlation features, and a subset of temporal change features. The temporal change feature subset is constructed by extracting the changing trend and fluctuation amplitude of the same indicator in different collection periods.

5. A chronic disease risk prediction system based on multimodal data fusion according to claim 1, characterized in that, The dynamic allocation logic for the weights of traditional Chinese medicine and Western medicine features in the fusion module is as follows: ； In the formula: The dynamic weight of the i-th feature; , , These are the weighting coefficients for discrimination, stability, and clinical relevance, respectively. , , The sum is 1; This is the average of the discrimination scores for all features; This is the average value of the stability of all features; This represents the average clinical relevance of all features; The total number of features participating in weight allocation; Let be the discriminative power of the i-th feature; The stability of the i-th feature; For the clinical relevance of feature i, Similarly; The hierarchical fusion is performed sequentially as follows: feature layer, feature set layer, and semantic layer. The feature layer performs standardization and normalization on individual features; the feature set layer performs weighted fusion on the subsets of Chinese and Western medicine features based on dynamic weights to obtain the feature set vector of Chinese and Western medicine. The semantic layer transforms the feature vectors of traditional Chinese medicine and Western medicine into feature vectors under a unified semantic space through a feature semantic mapping matrix, thereby completing the fusion.

6. The chronic disease risk prediction system based on multimodal data fusion according to claim 1, characterized in that, The prediction module's integrated traditional Chinese and Western medicine prediction logic is as follows: ； In the formula: Quantification of chronic disease risk levels; , , The integrated weights are based on fundamental risk, dialectical correction, and time series optimization. This represents the basic risk value for Western medicine. This is a correction value for TCM syndrome differentiation; This refers to the rate of change of time-series risk. The larger the value, the higher the risk; conversely, the smaller the value, the lower the risk. System users continuously monitor this data. The changing trend or presupposition and Risk prediction is conducted in this way.

7. A chronic disease risk prediction system based on multimodal data fusion according to claim 6, characterized in that, The , The calculation formula is: ； In the formula: This represents the total number of physicochemical indicators in Western medicine. Let be the weight of the j-th Western medicine physicochemical indicator; For indicator risk mapping function; This represents the total number of TCM syndrome characteristics; The weight of the k-th TCM syndrome feature; For syndrome differentiation risk function; in, >0, and ; >0, and .

8. A chronic disease risk prediction system based on multimodal data fusion according to claim 7, characterized in that, The expressions for the indicator risk mapping function and the syndrome differentiation risk function are as follows: ； In the formula: The risk amplification coefficient for the j-th indicator; The data is the normalized data for the j-th Western medicine physicochemical indicator; This represents the clinical standard reference value corresponding to the j-th indicator; The clinically permissible fluctuation range for the j-th indicator; Let be the nonlinear exponent of the j-th index; Let be the risk threshold parameter for the j-th indicator; This is the direction correction coefficient for the j-th index; This is a sign function; it returns 1 when the value inside the parentheses is positive, -1 when it is negative, and 0 when it is zero. The basic risk coefficient for the kth syndrome; is the standardized intensity value of the k-th TCM syndrome feature; Let be the intensity saturation coefficient of the k-th syndrome; Let be the intensity sensitivity coefficient of the k-th syndrome; This represents the number of other syndromes that have a synergistic effect with the k-th syndrome. Let be the co-risk coefficient between the k-th syndrome and the l-th co-syndrome; is the standardized intensity value of the l-th co-symptom.

9. A chronic disease risk prediction system based on multimodal data fusion according to claim 1, characterized in that, The upload module is interconnected with the processing module and the extraction module via a wireless network. The extraction module is interconnected with the fusion module via a wireless network. The fusion module is interconnected with the prediction module and the feedback module via a wireless network.

10. A method for predicting chronic disease risk based on multimodal data fusion, wherein the method is an implementation method of the chronic disease risk prediction system based on multimodal data fusion as described in any one of claims 1-9, characterized in that, include: Upload patient TCM syndrome description data and Western medicine physiochemical test data, store them according to data type, and establish a multi-source data index identifier containing fields such as patient unique identification code and data type identifier bit; Standardization processing, including semantic word segmentation, synonym mapping, and category labeling, is performed on TCM syndrome data. Outliers in Western medicine physicochemical data are filtered out using anomaly detection coefficients, and normalized multimodal data is output. Extract core syndrome factors from traditional Chinese medicine and key physicochemical indicators from Western medicine, calculate the correlation strength coefficient to screen effective correlation features, and construct a multi-dimensional feature set based on time-series change features; Based on feature discriminativeness, stability and clinical relevance, the weights of Chinese and Western medicine features are dynamically allocated. Hierarchical fusion is completed through feature layer normalization, feature set layer weighting and semantic layer mapping to generate a unified feature vector. The system sets up a predictive logic that integrates traditional Chinese and Western medicine, combining the basic risk value of Western medicine, the correction value of traditional Chinese medicine syndrome differentiation, and the time series risk change rate to quantify and extrapolate the risk level of chronic diseases. Receive the quantitative results of chronic disease risk levels and push the quantitative results to the designated receiving end.

Images