A health risk dynamic assessment method based on multi-modal fusion

Abstract

The application discloses a kind of health risk dynamic evaluation methods based on multi-modal fusion, it is related to health assessment technical field, including the following steps: obtaining and standardizing the multi-modal health data of to be evaluated object;Standardized health feature dataset is mapped;Multi-dimensional health feature set is initialized to dimension risk;Extract health change feature and carry out individual state correction, state phase division and phase migration analysis;Establish the matching relationship between phase change feature and health risk assessment dimension, generate individualized weight adjustment set;The initial health dimension weight is revised, and inertia constraint is carried out to revision process;Risk fusion calculation is carried out to initial health risk assessment set, and health risk dynamic evaluation result is generated.The application utilizes multi-modal health data fusion and health state phase analysis method, realizes health risk dynamic evaluation, has the advantages that evaluation continuity is strong, risk early warning accuracy is high and individual adaptability is strong.

Description

Technical Field

[0001] This invention relates to the field of health assessment technology, and in particular to a dynamic health risk assessment method based on multimodal fusion. Background Technology

[0002] With the development of digital health management and continuous health monitoring technologies, health risk assessment methods based on multi-source health data are gradually being applied to health management, chronic disease monitoring, and personalized health intervention scenarios. Existing technologies typically collect lifestyle data, physiological sign data, physical examination indicator data, and wearable device monitoring data to conduct risk analysis on users' health status and use risk scoring models to output corresponding health risk levels and early warning results.

[0003] Most existing health risk assessment methods employ static risk weights and fixed scoring rules, lacking dynamic correlation analysis of health change trends, risk state transitions, and individual differences within continuous time windows. This makes it difficult to reflect the changing state of health risk at different stages. Existing technologies have weak capabilities for dynamically adjusting weights between health risk dimensions and cannot adaptively correct risk assessment weights based on phase changes in health status. Consequently, the generated health risk assessment results lack continuity and dynamism, resulting in low accuracy in health risk warnings.

[0004] Therefore, how to provide a dynamic health risk assessment method based on multimodal fusion is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a dynamic health risk assessment method based on multimodal fusion. This invention utilizes multimodal health data fusion and health status phase analysis to achieve dynamic health risk assessment, which has the advantages of strong assessment continuity, high accuracy of risk warning, and strong individual adaptability.

[0006] A dynamic health risk assessment method based on multimodal fusion according to an embodiment of the present invention includes the following steps: Acquire multimodal health data of the object to be evaluated and perform standardization processing to generate a standardized health feature dataset; The standardized health feature dataset is feature-mapped according to the dimensions of health risk assessment to generate a multidimensional health feature set. Perform dimensional risk initialization processing on the multidimensional health feature set to generate an initial health risk assessment set; Based on the multidimensional health feature set, health change features within a continuous time window are extracted, and individual state correction, state phase division and phase migration analysis are performed to generate a set of health state phase trajectories. The phase change characteristics in the health status phase trajectory set are analyzed, and the phase change characteristics are matched with each health risk assessment dimension in the initial health risk assessment set to generate an individualized weight adjustment set. The initial health dimension weights in the initial health risk assessment set are corrected based on the individualized weight adjustment set, and the correction process is constrained by inertia to generate a dynamic health dimension weight set. The initial health risk assessment set is subjected to risk fusion calculation based on the dynamic health dimension weight set to generate dynamic health risk assessment results.

[0007] Optionally, the multimodal health data includes lifestyle data, traditional Chinese medicine constitution assessment data, meridian assessment data, physiological sign data, functional medicine health indicator data, and wearable device continuous monitoring data. The standardization process includes data cleaning, format unification, dimension normalization, timestamp alignment, and handling of missing and abnormal data.

[0008] Optionally, the generation of the multidimensional health feature set specifically includes: Based on the dimensions of lifestyle, traditional Chinese medicine constitution, meridian status, physiological signs, functional medicine indicators, and continuous monitoring trends, a dimensional correspondence between the standardized health feature dataset and the health risk assessment dimensions is established. Based on the dimensional correspondence, behavioral features are extracted from the lifestyle data in the standardized health feature dataset to generate lifestyle features. Based on the dimensional correspondence, the TCM constitution assessment data and meridian assessment data in the standardized health feature dataset are mapped to the constitution and meridian status to generate TCM constitution features and meridian status features. Based on the dimensional correspondence, the physiological signs data and functional medical health indicator data in the standardized health feature dataset are mapped to indicator intervals to generate physiological sign features and functional medical indicator features. Based on the continuous monitoring data of wearable devices in the standardized health feature dataset, the direction, magnitude and duration of indicator changes within the continuous time window are extracted to generate continuous monitoring trend features. By configuring corresponding feature source identifiers and feature time identifiers for lifestyle characteristics, TCM constitution characteristics, meridian status characteristics, physiological signs characteristics, functional medicine indicator characteristics, and continuous monitoring trend characteristics, a multidimensional health feature set is generated.

[0009] Optionally, the generation of the initial health risk assessment set specifically includes: The multidimensional health feature set is aggregated according to the health risk assessment dimensions, and dimension risk matching data is generated based on the normal reference range, risk classification interval and abnormal deviation direction of each aggregated health feature. Based on the dimensional risk matching data, the number of abnormal health features, the degree of abnormal deviation, and the duration of abnormality in each health risk assessment dimension are normalized and summarized to determine the initial dimensional risk value corresponding to each health risk assessment dimension. Based on the number of features, feature source identifiers, and feature time identifiers in the multidimensional health feature set, an initial weight allocation is performed for each health risk assessment dimension to determine the initial health dimension weights corresponding to each health risk assessment dimension. The initial dimension risk value and initial health dimension weight under the same health risk assessment dimension are configured as a set of dimension risk initialization records, and the risk initialization records of each dimension are encapsulated in a structured manner to generate an initial health risk assessment set.

[0010] Optionally, the generation of the health state phase trajectory set specifically includes: Based on the feature time identifiers in the multidimensional health feature set, the health features corresponding to each health risk assessment dimension are arranged in time series and segmented according to continuous time windows to generate a windowed health feature sequence. Based on the windowed health feature sequence, extract the direction of change, magnitude of change, duration of change and stability of change of indicators within each continuous time window to generate health change features; Based on the age, gender, body type, and past health status of the subjects to be evaluated, individual status corrections are made to the health change characteristics to generate individualized health change characteristics. Based on the dimensions of health risk assessment, the individualized health change characteristics are divided into state phases to determine the health state phase nodes corresponding to each continuous time window. Migration analysis is performed on health status phase nodes within adjacent consecutive time windows to determine the direction of phase change, the magnitude of phase migration, and the duration of phase dwell. Connect each health state phase node according to the characteristic time identifier, and configure the phase change direction, phase migration amplitude and phase dwell time between adjacent health state phase nodes to generate a set of health state phase trajectories.

[0011] Optionally, the generation of the individualized weight adjustment set specifically includes: Based on the health status phase trajectory set, the phase change features of each health risk assessment dimension within a continuous time window are extracted and matched with the dimension risk initialization records in the initial health risk assessment set to establish the phase dimension correlation relationship. Based on the correlation of phase dimensions, the risk change trend corresponding to the phase change direction is analyzed, and the risk change trend is converted into the weight adjustment direction of the corresponding health risk assessment dimension; Based on the phase migration amplitude, phase dwell time, initial dimension risk value, and initial health dimension weight, the degree of weight change of the corresponding health risk assessment dimension is quantified to determine the weight adjustment range; Based on the direction and magnitude of weight adjustment, the phase change characteristics of each health risk assessment dimension are used to determine the triggering conditions and configure the corresponding weight adjustment triggering identifiers. The direction, magnitude, and triggering identifier of weight adjustment for the same health risk assessment dimension are structured and encapsulated to generate an individualized weight adjustment set.

[0012] Optionally, the generation of the dynamic health dimension weight set specifically includes: The weight adjustment direction, weight adjustment magnitude, and weight adjustment trigger identifier in the individualized weight adjustment set are bound to the weights of the initial health dimensions in the initial health risk assessment set according to the health risk assessment dimensions, to generate a weight correction benchmark record; Based on the weight adjustment direction and magnitude in the weight adjustment benchmark record, the initial health dimension weights are directionally adjusted to generate candidate health dimension weights. Based on the weight changes of candidate health dimensions within the same health risk assessment dimension within a continuous time window, the direction of weight change, the magnitude of weight change, and the duration of weight retention are extracted to generate health weight inertia. The candidate health dimension weights are subjected to inertial constraint processing based on the health weight inertia, and the corrected health dimension weights are generated. Based on the weight adjustment trigger flag and the feature time flag corresponding to the continuous time window, configure the weight effective time and weight maintenance status for the corrected health dimension weights; The modified health dimension weights, weight effective time, and weight retention status corresponding to the same health risk assessment dimension are structured and encapsulated to generate a dynamic health dimension weight set.

[0013] Optionally, the generation of the dynamic health risk assessment results specifically includes: Align the corrected health dimension weights in the dynamic health dimension weight set with the initial dimension risk values ​​in the initial health risk assessment set according to the health risk assessment dimensions to generate a risk weight fusion record; Based on the risk weight fusion record, the initial dimension risk value and the corrected health dimension weight under the same health risk assessment dimension are weighted and fused to generate a dimension-weighted risk value. The dimension-weighted risk values ​​corresponding to each health risk assessment dimension are summarized and normalized according to the corrected sum of health dimension weights to determine the comprehensive health risk score. Based on the risk score range of the comprehensive health risk score, the health risk level of the subject to be assessed is determined, and abnormal health dimensions are screened based on the weighted risk values ​​of each dimension. Based on the weight maintenance status and corrected health dimension weights in the dynamic health dimension weight set, weight change results are generated, and the weight change results are correlated with abnormal health dimensions to generate risk change description data. The system generates early warning prompts based on comprehensive health risk scores, health risk levels, abnormal health dimensions, and risk change descriptions, and generates health risk warning information when the comprehensive health risk score reaches a preset warning threshold. The comprehensive health risk score, health risk level, abnormal health dimensions, weight change results, and health risk early warning information are structured and encapsulated to generate dynamic health risk assessment results.

[0014] The beneficial effects of this invention are: This invention integrates lifestyle data, traditional Chinese medicine constitution assessment data, meridian assessment data, physiological sign data, functional medicine health indicator data, and continuous monitoring data from wearable devices to establish a multidimensional set of health features covering multiple dimensions of health risk assessment. It also generates a set of health state phase trajectories by combining health change characteristics within a continuous time window. This allows the health risk assessment process to reflect not only the current health status of the assessed individual but also the dynamic changes in health risk within a continuous time window. Compared to existing risk scoring methods based solely on static indicators, this invention enables continuous analysis of health status migration trends, risk accumulation processes, and recovery processes, thereby improving the ability of health risk assessment results to represent the true health status.

[0015] This invention establishes a matching relationship between phase change characteristics and health risk assessment dimensions. It generates an individualized weight adjustment set based on the direction of phase change, the amplitude of phase migration, and the duration of phase dwell. Furthermore, it dynamically corrects the initial health dimension weights by incorporating health weight inertia, enabling the weights of different health risk assessment dimensions to dynamically adjust as the health status of the assessed individual changes. By introducing inertial constraints on the weight correction process based on the direction, amplitude, and duration of weight changes, the interference of short-term abnormal fluctuations on health risk assessment results can be reduced. This improves the stability and continuity of the dynamic health dimension weight set in continuous assessment scenarios, thereby enhancing the accuracy and reliability of the comprehensive health risk score.

[0016] This invention further performs risk fusion calculations on the initial health risk assessment set based on a dynamic health dimension weight set, generating a comprehensive health risk score, health risk level, abnormal health dimensions, weight change results, and health risk early warning information. It also generates corresponding early warning prompts by combining risk change description data. This ensures that the dynamic health risk assessment results not only output the risk level but also reflect the source of risk change, the state of risk enhancement, and the state of risk recovery. By dynamically comparing the comprehensive health risk score with a preset early warning threshold, it can promptly generate health risk early warning information when the health risk reaches the warning conditions, improving the timeliness and targeting of health risk early warnings. This is suitable for long-term continuous health management, chronic disease risk monitoring, and personalized health intervention scenarios. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a dynamic health risk assessment method based on multimodal fusion proposed in this invention; Figure 2 This is a flowchart illustrating the generation process of a health state phase trajectory set for a dynamic health risk assessment method based on multimodal fusion proposed in this invention. Figure 3 This is a flowchart illustrating the generation process of the dynamic health dimension weight set for a dynamic health risk assessment method based on multimodal fusion proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-3 A dynamic health risk assessment method based on multimodal fusion includes the following steps: Acquire multimodal health data of the object to be evaluated and perform standardization processing to generate a standardized health feature dataset; The standardized health feature dataset is feature-mapped according to the dimensions of health risk assessment to generate a multidimensional health feature set. Perform dimensional risk initialization processing on the multidimensional health feature set to generate an initial health risk assessment set; Based on the multidimensional health feature set, health change features within a continuous time window are extracted, and individual state correction, state phase division and phase migration analysis are performed to generate a set of health state phase trajectories. The phase change characteristics in the health status phase trajectory set are analyzed, and the phase change characteristics are matched with each health risk assessment dimension in the initial health risk assessment set to generate an individualized weight adjustment set. The initial health dimension weights in the initial health risk assessment set are corrected based on the individualized weight adjustment set, and the correction process is constrained by inertia to generate a dynamic health dimension weight set. The initial health risk assessment set is subjected to risk fusion calculation based on the dynamic health dimension weight set to generate dynamic health risk assessment results.

[0020] In this embodiment, multimodal health data includes lifestyle data, traditional Chinese medicine constitution assessment data, meridian assessment data, physiological sign data, functional medicine health indicator data, and wearable device continuous monitoring data. Standardization processing includes data cleaning, format unification, dimensional normalization, timestamp alignment, and handling of missing and abnormal data.

[0021] In this embodiment, the generation of the multidimensional health feature set specifically includes: Based on the dimensions of lifestyle, traditional Chinese medicine constitution, meridian status, physiological signs, functional medicine indicators, and continuous monitoring trends, a dimensional correspondence relationship is established between the standardized health feature dataset and the health risk assessment dimension. The dimensional correspondence relationship includes the data attribution relationship, dimension mapping relationship, and time window association relationship between each health data in the standardized health feature dataset and the corresponding health risk assessment dimension. Based on the dimensional correspondence, behavioral features are extracted from the lifestyle data in the standardized health feature dataset to generate lifestyle features, which include dietary regularity features, stable work and rest features, exercise frequency features, and stress behavior features. Based on the dimensional correspondence, the TCM constitution assessment data and meridian assessment data in the standardized health feature dataset are mapped to the constitution and meridian status to generate TCM constitution features and meridian status features. The TCM constitution features include Qi deficiency tendency features, Yang deficiency tendency features, phlegm-dampness tendency features, and constitution imbalance features. The meridian status features include meridian patency features, meridian blockage distribution features, left and right meridian balance features, and meridian status change trend features. Based on the dimensional correspondence, the physiological signs data and functional medicine health indicator data in the standardized health feature dataset are mapped to indicator intervals to generate physiological sign features and functional medicine indicator features. Physiological sign features include heart rate fluctuation features, blood pressure change features, weight change features, and sleep state features; functional medicine indicator features include metabolic abnormality features, nutritional status features, inflammatory response features, and functional indicator deviation features. Based on the continuous monitoring data of wearable devices in the standardized health feature dataset, the direction, magnitude, and duration of indicator changes within a continuous time window are extracted to generate continuous monitoring trend features; the continuous time window is obtained by dividing the time series according to the continuous distribution of timestamps in the continuous monitoring data of wearable devices. By configuring corresponding feature source identifiers and feature time identifiers for lifestyle characteristics, TCM constitution characteristics, meridian status characteristics, physiological signs characteristics, functional medicine indicator characteristics, and continuous monitoring trend characteristics, a multidimensional health feature set is generated.

[0022] In this embodiment, the generation of the initial health risk assessment set specifically includes: The multidimensional health feature set is aggregated according to the health risk assessment dimensions. Based on the normal reference range, risk classification interval, and abnormal deviation direction of each aggregated health feature, dimension risk matching data is generated. The normal reference range includes the normal value range, normal fluctuation range, and normal duration range of each health feature under the corresponding health risk assessment dimension. The risk classification interval includes the low-risk interval, medium-risk interval, and high-risk interval corresponding to each health feature. The abnormal deviation direction includes the upward deviation direction, downward deviation direction, and continuous fluctuation deviation direction of each health feature relative to the normal reference range. Based on the dimensional risk matching data, the number of abnormal health features, the degree of abnormal deviation, and the duration of abnormality in each health risk assessment dimension are normalized and summarized to determine the initial dimensional risk value for each health risk assessment dimension. The degree of abnormal deviation is the proportion of a health feature exceeding the normal reference range boundary to the width of the normal reference range; it is zero when it does not exceed the normal reference range. The duration of abnormality is the proportion of the number of consecutive time windows in which a health feature is in an abnormal deviation state to the total number of consecutive time windows; it is zero when no abnormal deviation state occurs. The normalization and summarization process is as follows: after normalizing the number of abnormal health features, the degree of abnormal deviation, and the duration of abnormality, a weighted sum is performed based on the risk sensitivity of the corresponding health risk assessment dimension. The risk sensitivity is the proportion of the number of health features falling into the high-risk range in the corresponding health risk assessment dimension to the total number of health features in the corresponding health risk assessment dimension; it is zero when no health features fall into the high-risk range. Based on the number of features, feature source identifiers, and feature time identifiers in the multidimensional health feature set, initial weights are assigned to each health risk assessment dimension to determine the initial health dimension weights corresponding to each health risk assessment dimension. The initial health dimension weights are determined by weighting the feature quantity weight, source credibility weight, and time continuity weight. Among them, the feature quantity weight is the proportion of the number of health features in the corresponding health risk assessment dimension to the total number of health features; the source credibility weight is the proportion of the sum of the source credibility levels of features in the corresponding health risk assessment dimension to the sum of the source credibility levels of all features; and the time continuity weight is the proportion of the number of time continuous windows of features in the corresponding health risk assessment dimension to the total number of time continuous windows of all features. The initial health dimension weight is the average of the feature quantity weight, source credibility weight, and time continuity weight. The initial dimension risk value and initial health dimension weight under the same health risk assessment dimension are configured as a set of dimension risk initialization records, and the risk initialization records of each dimension are encapsulated in a structured manner to generate an initial health risk assessment set.

[0023] In this embodiment, the generation of the health status phase trajectory set specifically includes: Based on the feature time identifiers in the multidimensional health feature set, the health features corresponding to each health risk assessment dimension are arranged in time series and segmented according to continuous time windows to generate a windowed health feature sequence. Based on the windowed health feature sequence, the direction, magnitude, duration, and stability of indicator changes within each consecutive time window are extracted to generate health change features. The direction of indicator change includes the increasing, decreasing, and stable directions of health features between adjacent consecutive time windows. The magnitude of change includes the amount of numerical change of health features between adjacent consecutive time windows, which is the proportion of the width of the corresponding health feature's normal reference range. The duration of change includes the number of consecutive time windows in which the health feature maintains the same direction of change. The stability of change includes the proportion of the number of fluctuations in health features within a consecutive time window to the total number of consecutive time windows. Based on the age, gender, body type, and past health status of the subjects to be evaluated, individual status corrections are made to the health change characteristics to generate individualized health change characteristics. Individual status correction includes: determining the age-appropriate range for the corresponding health risk assessment dimension based on age information; determining the gender-appropriate range for the corresponding health risk assessment dimension based on gender information; determining the constitution bias range corresponding to TCM constitution characteristics and meridian status characteristics based on constitution type; determining the historical risk benchmark for the corresponding health risk assessment dimension based on past health status; and calculating the degree of individual fit deviation of health change characteristics relative to the age-appropriate range, gender-appropriate range, constitution bias range, and historical risk benchmark. When the degree of individual fit deviation increases, the magnitude and duration of change of the corresponding health change characteristics are increased; when the degree of individual fit deviation decreases, the magnitude and duration of change of the corresponding health change characteristics are decreased; when the degree of individual fit deviation remains stable, the stability of the change of the corresponding health change characteristics is maintained, thus generating individualized health change characteristics. Based on the dimensions of health risk assessment, individualized health change characteristics are divided into state phases to determine the health state phase nodes corresponding to each continuous time window; the health state phase nodes include stable phase nodes, fluctuating phase nodes, risk accumulation phase nodes, and recovery phase nodes. The state phase division is determined based on the direction, magnitude, duration, and stability of the indicators in the individualized health change characteristics. Specifically: a stable phase node is defined as follows: when the indicator changes in a stable direction, the magnitude is within the corresponding normal fluctuation range, the duration is less than half the total number of consecutive time windows, and the stability is lower than the average stability of the corresponding changes in each consecutive time window under the same health risk assessment dimension; a fluctuating phase node is defined as follows: when the indicator changes alternately between increasing and decreasing directions, the magnitude is within the corresponding normal fluctuation range, and the stability is higher than the average stability of the corresponding changes in each consecutive time window under the same health risk assessment dimension; a risk accumulation phase node is defined as follows: when the indicator changes continuously in an increasing direction, the magnitude exceeds the width of the corresponding normal reference range, the duration is higher than half the total number of consecutive time windows, and the magnitude of the corresponding health risk assessment dimension within the consecutive time window continues to increase; and a recovery phase node is defined as follows: when the indicator changes from an increasing direction to a decreasing direction, the magnitude gradually decreases, the duration gradually shortens, and the magnitude of the corresponding health risk assessment dimension within the consecutive time window gradually decreases. Migration analysis is performed on health status phase nodes within adjacent consecutive time windows to determine the direction of phase change, the magnitude of phase migration, and the duration of phase dwell. The direction of phase change is determined based on the order of change of health state phase nodes between adjacent consecutive time windows; when the initial dimensional risk value corresponding to the later health state phase node is higher than that corresponding to the earlier health state phase node, it is determined to be the direction of risk increase; when the initial dimensional risk value corresponding to the later health state phase node is lower than that corresponding to the earlier health state phase node, it is determined to be the direction of risk decrease; when the initial dimensional risk value corresponding to the later health state phase node is the same as that corresponding to the earlier health state phase node, it is determined to be the direction of risk maintenance. Phase shift amplitude is the difference in the change amplitude of individualized health change characteristics corresponding to adjacent health state phase nodes, which accounts for the proportion of the width of the normal reference range of health characteristics under the corresponding health risk assessment dimension. Phase dwell time is the proportion of the number of consecutive time windows in which phase nodes of the same health state exist consecutively to the total number of consecutive time windows. Connect each health state phase node according to the characteristic time identifier, and configure the phase change direction, phase migration amplitude and phase dwell time between adjacent health state phase nodes to generate a set of health state phase trajectories.

[0024] In this embodiment, the generation of the individualized weight adjustment set specifically includes: Based on the health status phase trajectory set, the phase change features of each health risk assessment dimension within a continuous time window are extracted and matched with the dimension risk initialization records in the initial health risk assessment set to establish the phase dimension correlation; the phase change features include the phase change direction, phase dwell time and phase migration amplitude. The phase dimension correlation includes dimension identifier consistency relationship, feature source consistency relationship and time window overlap relationship; the dimension identifier consistency relationship is used to determine the health risk assessment dimension to which the phase change feature belongs; the feature source consistency relationship is used to determine the lifestyle feature, TCM constitution feature, meridian state feature, physiological sign feature, functional medicine indicator feature or continuous monitoring trend feature corresponding to the phase change feature; the time window overlap relationship is used to determine the correspondence between the phase change feature and the dimension risk initialization record within a continuous time window. Based on the correlation of phase dimensions, the risk change trend corresponding to the phase change direction is analyzed, and the risk change trend is converted into the weight adjustment direction of the corresponding health risk assessment dimension; Risk change trends include risk increase, risk decrease, and risk maintenance. When the phase change direction is risk increase, the weight adjustment direction of the corresponding health risk assessment dimension is determined as the weight increase direction; when the phase change direction is risk decrease, the weight adjustment direction of the corresponding health risk assessment dimension is determined as the weight decrease direction; when the phase change direction is risk maintenance, the weight adjustment direction of the corresponding health risk assessment dimension is determined as the weight maintenance direction. Based on the phase migration amplitude, phase dwell time, initial dimension risk value, and initial health dimension weight, the degree of weight change of the corresponding health risk assessment dimension is quantified to determine the weight adjustment range; The weight adjustment range is determined by normalizing and weighting the migration contribution, persistence contribution, and risk contribution, and then limiting the range in conjunction with the weight benchmark. The weight benchmark is the normalized value of the initial health dimension weights. Among them, the migration contribution is the normalized value of the phase migration amplitude, which is the proportion of the difference in the change amplitude of adjacent health state phase nodes to the width of the normal reference range; the persistence contribution is the weighted sum of the normalized value of the dwell time of the risk accumulation phase node and the inverse normalized value of the dwell time of the recovery phase node; the risk contribution is the weighted sum of the normalized value of the initial dimension risk value and the proportion of the number of high-risk health features. Based on the direction and magnitude of weight adjustment, trigger conditions are determined for the phase change characteristics of each health risk assessment dimension, and corresponding weight adjustment trigger flags are configured. Specifically: when the weight adjustment direction is upward and the weight adjustment magnitude is in the latter third of the range after sorting the weight adjustment magnitudes in the corresponding health risk assessment dimension from smallest to largest, a risk enhancement trigger flag is configured; when the weight adjustment direction is downward and the duration of the recovery phase node in the phase change characteristics is in the latter third of the range after sorting the duration of the recovery phase node in the corresponding health risk assessment dimension from smallest to largest, a recovery adjustment trigger flag is configured; when the weight adjustment direction is maintenance, or the weight adjustment magnitude is in the first third of the range after sorting the weight adjustment magnitudes in the corresponding health risk assessment dimension from smallest to largest, a weight maintenance trigger flag is configured. The direction, magnitude, and triggering identifier of weight adjustment for the same health risk assessment dimension are structured and encapsulated to generate an individualized weight adjustment set.

[0025] In this embodiment, the generation of the dynamic health dimension weight set specifically includes: The weight adjustment direction, weight adjustment magnitude, and weight adjustment trigger identifier in the individualized weight adjustment set are bound to the weights of the initial health dimensions in the initial health risk assessment set according to the health risk assessment dimensions, to generate a weight correction benchmark record; Based on the weight adjustment direction and magnitude in the weight adjustment benchmark record, the initial health dimension weights are directionally adjusted to generate candidate health dimension weights. The directional adjustment includes: when the weight adjustment direction is upward, the initial health dimension weights are added to the weight adjustment magnitude; when the weight adjustment direction is downward, the initial health dimension weights are subtracted from the weight adjustment magnitude; when the weight adjustment direction is to maintain the weights, the initial health dimension weights remain unchanged. Based on the weight changes of candidate health dimensions within the same health risk assessment dimension within a continuous time window, the direction of weight change, the magnitude of weight change, and the duration of weight retention are extracted to generate health weight inertia. The direction of weight change is determined based on the magnitude relationship between the weights of candidate health dimensions within adjacent consecutive time windows. When the weight of the candidate health dimension corresponding to the subsequent consecutive time window is greater than that corresponding to the previous consecutive time window, it is determined to be the direction of weight increase; when the weight of the candidate health dimension corresponding to the subsequent consecutive time window is less than that corresponding to the previous consecutive time window, it is determined to be the direction of weight decrease; when the weight of the candidate health dimension corresponding to the subsequent consecutive time window is equal to that corresponding to the previous consecutive time window, it is determined to be the direction of weight maintenance. The magnitude of weight change is the proportion of the difference in weights of candidate health dimensions within adjacent consecutive time windows to the total range of weights of candidate health dimensions under the corresponding health risk assessment dimension. The weight retention duration is the proportion of the number of consecutive time windows in which the weights of the candidate health dimensions are in the same direction of weight change to the total number of consecutive time windows. The candidate health dimension weights are subjected to inertia constraint processing based on the inertia of health weights to generate corrected health dimension weights. The inertia constraint processing includes: when the direction of weight change is consistent with the direction of weight adjustment, the proportion of the weight holding time to the total number of consecutive time windows is used as the inertia continuation ratio, the change in the candidate health dimension weight is multiplied by the inertia continuation ratio to obtain the inertia continuation amount, and the inertia continuation amount is added to the change in the candidate health dimension weight; when the direction of weight change is opposite to the direction of weight adjustment, the magnitude of weight change is multiplied by the change in the candidate health dimension weight to obtain the inertia reduction amount, and the inertia reduction amount is subtracted from the change in the candidate health dimension weight; when the weight holding time is within the first third of the interval after the weight holding time is sorted from smallest to largest under the same health risk assessment dimension, the average value of the change in the candidate health dimension weight within adjacent consecutive time windows under the same health risk assessment dimension is used as the maximum change amount in a single instance, and the change amount in the candidate health dimension weight is limited using the maximum change amount in a single instance as the upper limit of change; based on the changed amount of candidate health dimension weight after the limit processing, the candidate health dimension weights are updated to generate corrected health dimension weights. Based on the weight adjustment trigger flag and the feature time flag corresponding to the continuous time window, configure the weight effective time and weight maintenance status for the corrected health dimension weights; The configuration of the weight effective time includes: setting the start time of the corresponding continuous time window when the weight adjustment trigger flag first appears as the weight effective time of the corrected health dimension weight; when the weight adjustment trigger flag changes within the continuous time window, reconfiguring the start time of the corresponding continuous time window as the weight effective time of the corrected health dimension weight. Weight maintenance states include enhanced maintenance state, recovery maintenance state, and stable maintenance state; the risk enhancement trigger sign corresponds to the enhanced maintenance state, the recovery adjustment trigger sign corresponds to the recovery maintenance state, and the weight maintenance trigger sign corresponds to the stable maintenance state. The modified health dimension weights, weight effective time, and weight retention status corresponding to the same health risk assessment dimension are structured and encapsulated to generate a dynamic health dimension weight set.

[0026] In this embodiment, the generation of dynamic health risk assessment results specifically includes: Align the corrected health dimension weights in the dynamic health dimension weight set with the initial dimension risk values ​​in the initial health risk assessment set according to the health risk assessment dimensions to generate a risk weight fusion record; Based on the risk weight fusion record, the initial dimension risk value and the corrected health dimension weight under the same health risk assessment dimension are weighted and fused to generate a dimension-weighted risk value; the dimension-weighted risk value is the product of the initial dimension risk value and the corrected health dimension weight; when the same health risk assessment dimension corresponds to multiple consecutive time windows, the corrected health dimension weight that is currently in effect is selected to participate in the calculation according to the weight effective time. The weighted risk values ​​for each health risk assessment dimension are summarized and normalized according to the corrected sum of health dimension weights to determine the comprehensive health risk score. The comprehensive health risk score is obtained by summing the weighted risk values ​​of each dimension and dividing it by the sum of the corrected health dimension weights for each health risk assessment dimension. Based on the risk score range of the comprehensive health risk score, the health risk level of the subject to be assessed is determined, and abnormal health dimensions are screened based on the weighted risk values ​​of each dimension. The risk scoring intervals are divided according to the comprehensive health risk scores of the subjects to be assessed in the past, sorted from smallest to largest. The first third of the intervals are the intervals corresponding to low risk level, the middle third are the intervals corresponding to medium risk level, and the last third are the intervals corresponding to high risk level. Health risk levels include low risk level, medium risk level, and high risk level. Abnormal health dimension is a health risk assessment dimension whose dimension-weighted risk value is located in the last third of the interval after sorting the weighted risk values ​​of each dimension of the same object to be assessed from smallest to largest, and whose corresponding initial dimension risk value falls into the high-risk interval. Based on the weight maintenance status and corrected health dimension weights in the dynamic health dimension weight set, weight change results are generated, and these results are correlated with abnormal health dimensions to generate risk change description data. Weight change results include weight increase results, weight decrease results, and weight maintenance results. Enhanced maintenance status corresponds to weight increase results, restored maintenance status corresponds to weight decrease results, and stable maintenance status corresponds to weight maintenance results. Risk change description data includes abnormal health dimensions, corresponding weight change results, weight effective time, and corresponding dimension weighted risk value. The warning message is generated based on the comprehensive health risk score, health risk level, abnormal health dimensions and risk change description data. The health risk warning message is generated when the comprehensive health risk score reaches the preset warning threshold. The health risk warning message includes the identification of the object to be assessed, the warning trigger time and the warning message content. The warning trigger time is the start time of the corresponding continuous time window when the comprehensive health risk score first reaches the preset warning threshold; when the comprehensive health risk score reaches the preset warning threshold again, it is updated to the start time of the corresponding continuous time window when it reaches it again; the preset warning threshold is obtained by sorting the comprehensive health risk scores of the historical objects to be assessed from small to large, and taking the comprehensive health risk score corresponding to the starting position of the first three-quarters interval after sorting as the preset warning threshold. The warning prompts include abnormal health dimension prompts, health risk level prompts, and weight change prompts. Abnormal health dimension prompts are generated based on the name of the corresponding health risk assessment dimension, the corresponding weighted risk value, and the corresponding initial dimension risk value. When an abnormal health dimension corresponds to multiple health risk assessment dimensions, they are generated in descending order of the corresponding weighted risk values. Health risk level prompts are generated based on the health risk level and the comprehensive health risk score. When the comprehensive health risk score is in the upper half of the corresponding health risk level interval, a high-concern level prompt is generated; when the comprehensive health risk score is in the lower half of the corresponding health risk level interval, a basic level prompt is generated. Weight change prompts are generated based on the weight change result, the weight's effective time, and the corresponding revised health dimension weight. When the weight change result is an increase, a risk enhancement prompt is generated; when the weight change result is a decrease, a risk recovery prompt is generated; when the weight change result is the same, a risk stabilization prompt is generated. The comprehensive health risk score, health risk level, abnormal health dimensions, weight change results, and health risk early warning information are structured and encapsulated to generate dynamic health risk assessment results.

[0027] Example 1: To verify the feasibility of this invention in practice, it was applied to a continuous health management scenario in a health management institution. The subjects to be assessed were individuals with long-term irregular work and rest schedules, unbalanced diets, insufficient exercise, and a tendency towards chronic metabolic abnormalities. Existing health risk assessment methods mostly rely on periodic physical examination indicators, physiological parameters, or single health questionnaire results for risk judgment. They lack the ability to dynamically analyze the changes in health status within a continuous time window. They struggle to accurately identify the fluctuations, accumulations, and recovery states of health risks during continuous monitoring, easily leading to problems such as delayed risk identification, misjudgments of abnormal fluctuations, and untimely updates to risk levels.

[0028] When applying this invention, the system first acquires lifestyle data, traditional Chinese medicine constitution assessment data, meridian assessment data, physiological sign data, functional medicine health indicator data, and continuous monitoring data from wearable devices for the subject to be assessed. Data from different sources is then processed for format standardization, dimension normalization, timestamp alignment, and handling of missing and abnormal data to generate a standardized health feature dataset. Subsequently, a multidimensional health feature set is established based on health risk assessment dimensions, and health change features are generated by combining the direction, magnitude, duration, and stability of indicator changes within a continuous time window. The system further performs individual state correction on the health change features based on the subject's age, gender, constitution type, and past health status, and then generates a set of health state phase trajectories by combining state phase division and phase migration analysis.

[0029] During continuous health monitoring, the system generates an individualized weight adjustment set based on the phase change direction, phase migration amplitude, and phase dwell time in the health status phase trajectory set, and dynamically corrects the initial health dimension weights by incorporating health weight inertia. When the assessed object continuously exhibits risk accumulation phase nodes within a continuous time window, the system increases the dynamic health dimension weight of the corresponding health risk assessment dimension; when recovery phase nodes persist, the system decreases the dynamic health dimension weight of the corresponding health risk assessment dimension; when abnormal fluctuations occur within a short period, the system uses an inertial constraint mechanism to limit the abrupt change in health dimension weights, thereby reducing the impact of occasional abnormal data on the comprehensive health risk score. The system further performs risk fusion calculations on the initial health risk assessment set based on the dynamic health dimension weight set, generating a comprehensive health risk score, health risk level, abnormal health dimensions, and health risk warning information. This ensures that the dynamic health risk assessment results reflect the risk change trend of the assessed object during continuous monitoring, improving the continuity, stability, and accuracy of risk warnings in the health risk assessment results.

[0030] Table 1. Performance Comparison of the Invention and Traditional Dynamic Health Risk Assessment Methods

[0031] As can be clearly seen from Table 1, the method of the present invention is superior to the traditional method in many indicators.

[0032] The method of this invention achieves a health risk identification accuracy of 92.1%, higher than the 86.4% of traditional static assessment methods. This is because the invention integrates lifestyle data, traditional Chinese medicine constitution assessment data, meridian assessment data, physiological sign data, and continuous monitoring data from wearable devices to establish a multidimensional health characteristic set. It then combines this with dynamic analysis of the phase change process of health status within a continuous time window, enabling the health risk assessment results to reflect the actual health changes of the assessed individual at different stages, thereby improving the accuracy of health risk identification.

[0033] Regarding the accuracy of identifying continuous risk changes, the method of this invention achieves 88.6%, while the traditional static assessment method achieves 79.8%. This invention constructs a set of health state phase trajectories, performs migration analysis on risk accumulation phase nodes, recovery phase nodes, and fluctuation phase nodes, and generates an individualized weight adjustment set by combining the phase change direction, phase migration amplitude, and phase dwell time. This enables the system to identify risk change trends within a continuous time window, thus improving the accuracy of continuous risk change identification.

[0034] In terms of dynamic weight adjustment response time, the method of this invention is 4.1s, while the traditional static evaluation method is 5.8s. This invention applies inertial constraints to the candidate health dimension weights through health weight inertia and dynamically updates the health dimension weights according to the weight adjustment trigger flag, reducing the processing delay caused by repeated weight calculations in the traditional static method, thus shortening the dynamic weight adjustment response time.

[0035] Regarding the volatility of risk scores, the method of this invention achieves 7.9%, lower than the 12.5% ​​of traditional static assessment methods. This invention limits short-term abnormal fluctuations by introducing weight retention duration, inertia continuation ratio, and inertia reduction amount. Furthermore, it reduces the impact of occasional abnormal data on the comprehensive health risk score through a dynamic health dimension weight correction process within a continuous time window. As a result, the volatility of risk scores is reduced, and the stability of continuous monitoring is improved from 84.1% to 90.3%.

[0036] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for dynamic health risk assessment based on multi-modal fusion, characterized in that, Includes the following steps: Acquire multimodal health data of the object to be evaluated and perform standardization processing to generate a standardized health feature dataset; The standardized health feature dataset is feature-mapped according to the dimensions of health risk assessment to generate a multidimensional health feature set. Perform dimensional risk initialization processing on the multidimensional health feature set to generate an initial health risk assessment set; Based on the multidimensional health feature set, health change features within a continuous time window are extracted, and individual state correction, state phase division and phase migration analysis are performed to generate a set of health state phase trajectories. The phase change characteristics in the health status phase trajectory set are analyzed, and the phase change characteristics are matched with each health risk assessment dimension in the initial health risk assessment set to generate an individualized weight adjustment set. The initial health dimension weights in the initial health risk assessment set are corrected based on the individualized weight adjustment set, and the correction process is constrained by inertia to generate a dynamic health dimension weight set. The initial health risk assessment set is subjected to risk fusion calculation based on the dynamic health dimension weight set to generate dynamic health risk assessment results.

2. The health risk dynamic assessment method based on multi-modal fusion according to claim 1, characterized in that, The multimodal health data includes lifestyle data, traditional Chinese medicine constitution assessment data, meridian assessment data, physiological sign data, functional medicine health indicator data, and wearable device continuous monitoring data. The standardization process includes data cleaning, format unification, dimension normalization, timestamp alignment, and handling of missing and abnormal data.

3. The health risk dynamic assessment method based on multi-modal fusion according to claim 1, characterized in that, The generation of the multidimensional health feature set specifically includes: Based on the dimensions of lifestyle, traditional Chinese medicine constitution, meridian status, physiological signs, functional medicine indicators, and continuous monitoring trends, a dimensional correspondence between the standardized health feature dataset and the health risk assessment dimensions is established. Based on the dimensional correspondence, behavioral features are extracted from the lifestyle data in the standardized health feature dataset to generate lifestyle features. Based on the dimensional correspondence, the TCM constitution assessment data and meridian assessment data in the standardized health feature dataset are mapped to the constitution and meridian status to generate TCM constitution features and meridian status features. Based on the dimensional correspondence, the physiological signs data and functional medical health indicator data in the standardized health feature dataset are mapped to indicator intervals to generate physiological sign features and functional medical indicator features. Based on the continuous monitoring data of wearable devices in the standardized health feature dataset, the direction, magnitude and duration of indicator changes within the continuous time window are extracted to generate continuous monitoring trend features. By configuring corresponding feature source identifiers and feature time identifiers for lifestyle characteristics, TCM constitution characteristics, meridian status characteristics, physiological signs characteristics, functional medicine indicator characteristics, and continuous monitoring trend characteristics, a multidimensional health feature set is generated.

4. The health risk dynamic assessment method based on multi-modal fusion according to claim 1, characterized in that, The generation of the initial health risk assessment set specifically includes: The multidimensional health feature set is aggregated according to the health risk assessment dimensions, and dimension risk matching data is generated based on the normal reference range, risk classification interval and abnormal deviation direction of each aggregated health feature. Based on the dimensional risk matching data, the number of abnormal health features, the degree of abnormal deviation, and the duration of abnormality in each health risk assessment dimension are normalized and summarized to determine the initial dimensional risk value corresponding to each health risk assessment dimension. Based on the number of features, feature source identifiers, and feature time identifiers in the multidimensional health feature set, an initial weight allocation is performed for each health risk assessment dimension to determine the initial health dimension weights corresponding to each health risk assessment dimension. The initial dimension risk value and initial health dimension weight under the same health risk assessment dimension are configured as a set of dimension risk initialization records, and the risk initialization records of each dimension are encapsulated in a structured manner to generate an initial health risk assessment set.

5. The health risk dynamic assessment method based on multi-modal fusion according to claim 1, characterized in that, The generation of the health state phase trajectory set specifically includes: Based on the feature time identifiers in the multidimensional health feature set, the health features corresponding to each health risk assessment dimension are arranged in time series and segmented according to continuous time windows to generate a windowed health feature sequence. Based on the windowed health feature sequence, extract the direction of change, magnitude of change, duration of change and stability of change of indicators within each continuous time window to generate health change features; Based on the age, gender, body type, and past health status of the subjects to be evaluated, individual status corrections are made to the health change characteristics to generate individualized health change characteristics. Based on the dimensions of health risk assessment, the individualized health change characteristics are divided into state phases to determine the health state phase nodes corresponding to each continuous time window. Migration analysis is performed on health status phase nodes within adjacent consecutive time windows to determine the direction of phase change, the magnitude of phase migration, and the duration of phase dwell. Connect each health state phase node according to the characteristic time identifier, and configure the phase change direction, phase migration amplitude and phase dwell time between adjacent health state phase nodes to generate a set of health state phase trajectories.

6. The health risk dynamic assessment method based on multi-modal fusion according to claim 1, characterized in that, The generation of the individualized weight adjustment set specifically includes: Based on the health status phase trajectory set, the phase change features of each health risk assessment dimension within a continuous time window are extracted and matched with the dimension risk initialization records in the initial health risk assessment set to establish the phase dimension correlation relationship. Based on the correlation of phase dimensions, the risk change trend corresponding to the phase change direction is analyzed, and the risk change trend is converted into the weight adjustment direction of the corresponding health risk assessment dimension; Based on the phase migration amplitude, phase dwell time, initial dimension risk value, and initial health dimension weight, the degree of weight change of the corresponding health risk assessment dimension is quantified to determine the weight adjustment range; Based on the direction and magnitude of weight adjustment, the phase change characteristics of each health risk assessment dimension are used to determine the triggering conditions and configure the corresponding weight adjustment triggering identifiers. The direction, magnitude, and triggering identifier of weight adjustment for the same health risk assessment dimension are structured and encapsulated to generate an individualized weight adjustment set.

7. The health risk dynamic assessment method based on multi-modal fusion according to claim 1, characterized in that, The generation of the dynamic health dimension weight set specifically includes: The weight adjustment direction, weight adjustment magnitude, and weight adjustment trigger identifier in the individualized weight adjustment set are bound to the weights of the initial health dimensions in the initial health risk assessment set according to the health risk assessment dimensions, to generate a weight correction benchmark record; Based on the weight adjustment direction and magnitude in the weight adjustment benchmark record, the initial health dimension weights are directionally adjusted to generate candidate health dimension weights. Based on the weight changes of candidate health dimensions within the same health risk assessment dimension within a continuous time window, the direction of weight change, the magnitude of weight change, and the duration of weight retention are extracted to generate health weight inertia. The candidate health dimension weights are subjected to inertial constraint processing based on the health weight inertia, and the corrected health dimension weights are generated. Based on the weight adjustment trigger flag and the feature time flag corresponding to the continuous time window, configure the weight effective time and weight maintenance status for the corrected health dimension weights; The modified health dimension weights, weight effective time, and weight retention status corresponding to the same health risk assessment dimension are structured and encapsulated to generate a dynamic health dimension weight set.

8. The health risk dynamic assessment method based on multi-modal fusion according to claim 1, characterized in that, The generation of the dynamic health risk assessment results specifically includes: Align the corrected health dimension weights in the dynamic health dimension weight set with the initial dimension risk values ​​in the initial health risk assessment set according to the health risk assessment dimensions to generate a risk weight fusion record; Based on the risk weight fusion record, the initial dimension risk value and the corrected health dimension weight under the same health risk assessment dimension are weighted and fused to generate a dimension-weighted risk value. The dimension-weighted risk values ​​corresponding to each health risk assessment dimension are summarized and normalized according to the corrected sum of health dimension weights to determine the comprehensive health risk score. Based on the risk score range of the comprehensive health risk score, the health risk level of the subject to be assessed is determined, and abnormal health dimensions are screened based on the weighted risk values ​​of each dimension. Based on the weight maintenance status and corrected health dimension weights in the dynamic health dimension weight set, weight change results are generated, and the weight change results are correlated with abnormal health dimensions to generate risk change description data. The system generates early warning prompts based on comprehensive health risk scores, health risk levels, abnormal health dimensions, and risk change descriptions, and generates health risk warning information when the comprehensive health risk score reaches a preset warning threshold. The comprehensive health risk score, health risk level, abnormal health dimensions, weight change results, and health risk early warning information are structured and encapsulated to generate dynamic health risk assessment results.

Images