Thyroid whole-cycle health management method

Abstract

The invention relates to the field of medical health management, and discloses a thyroid-oriented whole-cycle health management method, which comprises the following steps of: firstly obtaining thyroid multi-source information data of a target population to form a data set, then extracting fusion features to generate a time sequence feature vector, and analyzing and calculating occurrence risk factors and scores of feature dimensions in a preset time window according to the time sequence feature vector; a comprehensive risk score is obtained through time sequence analysis, risk grades are divided, high-risk individuals are marked, and a personalized screening strategy is generated; then combining historical data of high-risk individuals, analyzing key driving paths with high risk scores, associating an external knowledge base to predict recurrence frequency, and finally dynamically adjusting and optimizing screening and intervention strategies according to the recurrence frequency; precise evaluation, grading and personalized management of individual thyroid health risks are achieved, then risk key driving paths are found out, recurrence is predicted, and the strategy is dynamically optimized to improve the disease prevention and treatment effect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical health management, and more particularly to a thyroid health management method for the whole cycle. BACKGROUND

[0002] At present, the technical development of thyroid health management mainly revolves around several core directions. At the data level, research and practice focus on collecting multi-source information such as clinical indicators, behaviors and environments from electronic medical records, medical images and wearable devices. In terms of analysis methods, statistical models and machine learning algorithms are widely used to stratify individuals based on historical data and identify key risk factors related to thyroid diseases. At the application level, existing clinical decision support systems mainly rely on authoritative clinical guidelines to provide standardized screening recommendations and intervention strategies for individuals of different risk levels. These technologies collectively form the current foundation framework for thyroid health management, aiming to achieve systematic monitoring and management of thyroid diseases through data-driven risk assessment and standardized processes.

[0003] However, the existing technology still has the following disadvantages: Firstly, the existing technology relies on single static data, making it difficult to integrate multi-source heterogeneous information and capture the dynamic evolution of health status; Secondly, the existing technology mostly stays at correlation analysis, failing to reveal the deep causal mechanism leading to increased risk, making the intervention lack of pertinence; Thirdly, the existing technology mostly adopts standardized "one-size-fits-all" strategies, failing to dynamically and prospectively adjust according to individual unique etiological pathways and future recurrence risks. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides a thyroid health management method for the whole cycle to solve the problems existing in the background art.

[0005] The present application provides the following technical solution: a thyroid health management method for the whole cycle, comprising: S1: acquiring thyroid multi-source information data of a target population and forming a multi-source information data set; S2: performing feature extraction and fusion on the multi-source information data set to generate a time series feature vector representing the individual's thyroid health status; S3: analyzing and calculating the risk factors and risk scores of each feature dimension on the occurrence of thyroid diseases within a preset time window according to the generated time series feature vector; S4: performing time series analysis on the calculated risk factor of occurrence and risk score of occurrence to obtain a comprehensive risk score, based on the comprehensive risk score, dividing each individual in the target population into a risk level, marking the divided high-risk level individuals, and generating an individualized screening strategy for each individual, the screening strategy including recommended screening index combination and screening frequency; S5: according to the divided high-risk level individuals, combining the historical thyroid multi-source information data of high-risk individuals, analyzing the key driving path that causes the risk score to continue to rise; S6: correlating the analyzed key driving path with the external knowledge base to predict the recurrence frequency of the key driving path; S7: according to the recurrence frequency of the key driving path, dynamically adjusting and optimizing the screening strategy and the intervention strategy corresponding to the key driving path.

[0006] Preferably, the S1 obtains multi-source information data of thyroid patients from multiple data sources through a multi-modal perception terminal, wherein the multi-source information data includes clinical data, behavior data, environmental data and psychological data; A cross-modal feature alignment network is used to map the multi-source heterogeneous data to a unified semantic space to generate a high-dimensional multi-source information data set.

[0007] Preferably, the S2 includes feature extraction and fusion of the multi-source information data set: The thyroid function index imageomics features and pathological section information in the clinical data are processed by a deep time series convolution network to extract the dynamic pattern in the clinical data over time and generate a first time series feature vector; wherein the deep time series convolution network is stacked by at least two residual convolution blocks, each residual convolution block contains a one-dimensional convolution layer, an activation function layer and a hollow convolution layer, which is used to capture the evolution dynamic pattern of different time scales in the clinical data; The medication adherence, exercise mode in the behavior data and regional iodine exposure level in the environmental data are used to construct an association graph, and a graph neural network is used to process the association graph to mine the non-linear association between the behavior data and the environmental data, and generate a graph structure feature vector; wherein the construction of the association graph includes: taking medication adherence, exercise mode and regional iodine exposure level as nodes of the graph, and determining the weight of the edge between the nodes according to the preset medical rules or the Pearson correlation coefficient between the data; The emotion fluctuation and stress index in the psychological data are processed by a long short-term memory network to capture the long time series dependence relationship in the psychological data to generate a second time series feature vector; The extracted first time sequence feature vector, the graph structure feature vector and the second time sequence feature vector are fused by a cross-modal attention fusion mechanism to generate a time sequence feature vector representing the health status of the individual thyroid.

[0008] Preferably, the S3 assigns weights to each dimension of the time sequence feature vector within a preset time window through a time sequence attention network to quantify the differential effects of clinical indicator fluctuations, behavior pattern changes, environmental exposure accumulations and psychological state evolution on the thyroid health status at different time points, thereby dynamically identifying key risk factors. The weighted feature vector is input into a risk prediction model trained on a disease-specific data set containing thyroid function indicators, imaging features and outcome labels, which is used to calculate the risk score of thyroid disease occurrence at a specific future time point; A model explanation method based on game theory is used to reverse analyze the decision-making process of the risk prediction model, decompose the risk score into the contribution of each feature dimension, and generate an individualized risk factor contribution map.

[0009] Preferably, the S4 performs time sequence trend analysis on multiple occurrence risk scores within a preset historical time window, calculates the risk change rate, and combines the current occurrence risk score and occurrence risk factors to calculate a comprehensive risk score by weighted summation; The comprehensive risk score is compared with a preset risk level threshold interval, each individual is divided into a low risk, medium risk or high risk level, and high-risk individuals are marked; For each individual, a personalized screening strategy is generated based on its risk level and the generated individualized risk factor contribution map, and the generation of the screening strategy includes: When the individual is classified as a high-risk level, the top N risk factors corresponding to the screening indicator combination with the highest contribution are recommended, and a first screening frequency is set; When the individual is classified as a medium-risk level, a standard screening indicator combination is recommended, and a second screening frequency lower than the first screening frequency is set; When the individual is classified as a low-risk level, a basic screening indicator combination is recommended, and a third screening frequency lower than the second screening frequency is set.

[0010] Preferably, the S5 constructs a directed acyclic graph including each feature variable in the historical thyroid multi-source information data of the high-risk level individual as a causal hypothesis graph, wherein the nodes of the directed acyclic graph represent the feature variables, and the directed edges between the nodes represent the preset causal relationship. A causal effect estimation model based on the backdoor criterion is adopted to identify all backdoor paths between any two target variables on the causal hypothesis graph, and the causal effect value represented by the directed edge is calculated by adjusting the set of confounding variables. Traverse all complete paths from the root node with no incoming edges to the target node representing the occurrence of the risk score in the directed acyclic graph, and multiply the average causal effect values ​​of each directed edge on each path to obtain the total causal effect value of the path. Based on the calculated total causal effect value of each path, the paths are sorted in descending order, and the path with the highest total causal effect value is identified and output as the key driving path that causes the risk score of the high-risk individual to continue to rise.

[0011] Preferably, in step S6, the nodes and directed edges in the key driving path are semantically matched and mapped with the standardized medical ontology and causal associations pre-stored in the external knowledge base, so as to identify the standardized causal path corresponding to the key driving path in the external knowledge base. From the external knowledge base, retrieve multiple historical case data associated with the standardized causal path, and extract the recurrence event records of the corresponding path in each historical case; Based on the historical case data, a survival analysis model is constructed, using the path length, node type, and causal effect value of each directed edge of the key driving path as covariates, and the recurrence event of the path as the endpoint event, to calculate the recurrence probability of the key driving path within a future preset time window. The calculated recurrence probability is output as the predicted recurrence frequency value of the key driving path.

[0012] Preferably, in step S7, the predicted recurrence frequency of the key driving path is compared with a preset frequency threshold, and each key driving path is divided into a high recurrence risk path or a low recurrence risk path. For the high recurrence risk pathway, the corresponding screening and intervention strategies are strengthened and adjusted. The strengthening and adjustment of the screening strategy includes: adding specific screening indicators for key nodes of the pathway and increasing the screening frequency, based on the original individualized screening strategy. The strengthening and adjustment of the intervention strategy includes: querying multiple intervention measures matching the node type of the high recurrence risk pathway from a preset intervention strategy knowledge base, and weighting and ranking the intervention measures according to their pathway suppression success rates recorded in historical cases to generate a combined intervention plan containing multiple intervention measures. For the aforementioned low recurrence risk pathways, maintain or reduce the corresponding screening frequency and adopt standardized intervention protocols based on clinical guidelines; By integrating all key driver-adjusted screening and intervention strategies, a personalized health management plan is generated. Finally, the generated personalized health management plan and recurrence frequency prediction value are output and sent to the target terminal via wireless communication and displayed in a visual interface. The target terminal includes, but is not limited to, doctor workstations and patient mobile terminal devices, for doctors and patients to view and execute.

[0013] The technical effects and advantages of this invention are as follows: By integrating multi-source heterogeneous data and constructing time-series feature vectors, comprehensive and dynamic monitoring of individual health status is achieved, enabling earlier and more accurate capture of early signals and evolution trends of diseases.

[0014] By combining the high-risk individuals with their historical multi-source thyroid information data, the key driving pathways leading to the continuous rise in their risk scores were analyzed. This represents a cognitive leap from surface analysis to in-depth analysis, enabling intervention measures to directly target the core causes and greatly improving the accuracy and effectiveness of treatment.

[0015] By dynamically optimizing management strategies through predicting relapse frequency, health management is upgraded from a passive, standardized response to a proactive, adaptive, and precise intervention based on individual future risk prediction, thus achieving truly personalized management throughout the entire lifecycle. Attached Figure Description

[0016] Figure 1 This is a diagram illustrating the method steps of the present invention.

[0017] Figure 2 This is a system structure block diagram of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The method for whole-cycle thyroid health management involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 The embodiment shown provides a method for full-cycle thyroid health management, including: S1: Obtain multi-source information data on the thyroid gland of the target population and form a multi-source information dataset.

[0020] In this embodiment, S1 obtains multi-source information data of thyroid patients from multiple data sources through a multimodal sensing terminal, wherein the multi-source information data includes clinical data, behavioral data, environmental data, and psychological data; A cross-modal feature alignment network is used to map the multi-source heterogeneous data to a unified semantic space to generate a high-dimensional multi-source information dataset.

[0021] It should be specifically noted that data is acquired from multiple data sources through a multimodal sensing terminal. Clinical data can originate from electronic medical records in the hospital information system, specifically including time-series test results of thyroid function indicators such as TSH, FT3, and FT4, as well as DICOM files of thyroid ultrasound images. Behavioral data can be acquired from smart bracelets worn by patients, including time-series data such as daily steps, sleep duration, and heart rate variability. Environmental data can be obtained based on the geographical coordinates of the patient's residence by calling public environmental database APIs, such as regional average iodine content in drinking water and annual average PM2.5 concentration. Psychological data can be quantified through scores on the PHQ-9 depression scale and GAD-7 anxiety scale, which are periodically filled out by patients via a mobile app. Subsequently, a cross-modal feature alignment network is used for processing. This network includes feature extractors for different data types: for example, a one-dimensional convolutional neural network is used to extract time-series features of clinical test indicators, a convolutional neural network is used to extract radiomics features of ultrasound images, a long short-term memory network is used to extract periodic features of behavioral data, and a multilayer perceptron is used to process environmental and psychological scale data. These feature extractors map their respective output feature vectors to a unified 256-dimensional semantic space through a shared projection layer, and are trained using a contrastive learning loss function to make the representation vectors of different modalities of the same patient as close as possible in the semantic space, thereby ultimately generating a high-dimensional multi-source information dataset that can comprehensively and deeply integrate individual health information.

[0022] S2: Extract and fuse features from the multi-source information dataset to generate a time-series feature vector representing the individual's thyroid health status.

[0023] In this embodiment, S2 performing feature extraction and fusion on the multi-source information dataset includes: The radiomics features of thyroid function indicators and pathological slide information in the clinical data are processed using a deep temporal convolutional network to extract the dynamic patterns of evolution over time in the clinical data and generate a first temporal feature vector. The deep temporal convolutional network is composed of at least two stacked residual convolutional blocks. Each residual convolutional block contains a one-dimensional convolutional layer, an activation function layer and a dilated convolutional layer, which are used to capture the dynamic patterns of evolution at different time scales in the clinical data. A correlation graph is constructed based on medication adherence and exercise patterns in the behavioral data and regional iodine exposure levels in the environmental data. A graph neural network is then used to process the correlation graph to uncover the nonlinear correlation between the behavioral data and the environmental data, generating a graph structure feature vector. The construction of the correlation graph includes using medication adherence, exercise patterns, and regional iodine exposure levels as nodes in the graph, and determining the weights of the edges between nodes based on preset medical rules or the Pearson correlation coefficient between the data. The emotional fluctuations and stress index in the psychological data are processed using a long short-term memory network to capture long-term temporal dependencies in the psychological data and generate a second temporal feature vector. By employing a cross-modal attention fusion mechanism, the extracted first temporal feature vector, graph structure feature vector, and second temporal feature vector are weighted and fused to generate a temporal feature vector representing an individual's thyroid health status.

[0024] Specifically, for clinical data, a 3-channel time-series input deep temporal convolutional network was constructed using a patient's TSH, FT3, and FT4 test values ​​over 12 consecutive months. This network processed the data using stacked residual convolutional blocks (with dilated convolutions having a dilation rate of 1, 2, and 4 to capture weekly, monthly, and quarterly patterns), outputting a 64-dimensional first temporal feature vector that encodes the dynamic evolution of the patient's thyroid function. For behavioral and environmental data, a correlation graph was constructed containing three nodes: "medication adherence (score)," "average weekly exercise time (hours)," and "regional iodine exposure level (ug / L)." The edge weights were determined based on the Pearson correlation coefficient of the historical data for these three nodes. This graph was then processed by a graph convolutional network, outputting a 32-dimensional graph structure feature vector that quantifies the complex interactions between behavioral and environmental factors. The patient's daily stress index (1-10 points) recorded via an app was input into an LSTM network, whose hidden states captured long-term emotional fluctuations, ultimately outputting a 32-dimensional second temporal feature vector. Finally, the three feature vectors (64-dimensional, 32-dimensional, and 32-dimensional) are input into a cross-modal attention fusion mechanism. This mechanism calculates the attention weights between different modal features and assigns higher weights to the modality most relevant to the current health status. For example, when clinical indicators are abnormal, the weight of the first temporal feature vector will increase significantly. Finally, the weighted fusion generates a 128-dimensional temporal feature vector that can comprehensively represent the individual's current thyroid health status.

[0025] S3: Based on the generated time-series feature vector, analyze and calculate the risk factors and risk scores for the occurrence of thyroid diseases for each feature dimension within a preset time window.

[0026] In this embodiment, S3 assigns weights to each dimension of the temporal feature vector through a temporal attention network within a preset time window to quantify the differential impact of fluctuations in clinical indicators, changes in behavioral patterns, cumulative environmental exposure, and evolution of psychological state at different time points on thyroid health status, thereby dynamically identifying key risk factors. The weighted feature vector is input into the risk prediction model, which is trained on a disease-specific dataset containing thyroid function indicators, imaging features and outcome labels, and is used to calculate the risk score of thyroid disease occurrence at a specific future time point. A game theory-based model interpretation method is used to reverse analyze the decision-making process of the risk prediction model, deconstructing the risk score into the contribution of each feature dimension to generate an individualized risk factor contribution map.

[0027] It should be specifically explained that the 128-dimensional temporal feature vector generated by S2 is input into a temporal attention network. This network calculates the attention weight of the feature vector for each month over the past 12 months. For example, if a patient's thyroid function indicators have fluctuated dramatically in the last 3 months, the feature vectors for these 3 months will be assigned a higher weight (e.g., 0.4), while the weights for earlier stable months will be lower (e.g., 0.05), thus dynamically identifying "drastic fluctuations in recent clinical indicators" as a key risk factor. This weighted feature vector is then input into an XGBoost risk prediction model pre-trained on a disease-specific dataset containing thyroid function, ultrasound images, and final diagnostic outcomes of tens of thousands of patients. The model outputs a value between 0 and 1 as the patient's risk score for developing thyroid nodules in the next 6 months, for example, 0.75. Finally, the SHapley Additive exPlanations (SHAP) method was used to reverse-analyze the decision of the XGBoost model, calculating the contribution of each feature dimension (such as "TSH level" and "sleep irregularity score") to the final risk score of 0.75. This resulted in an individualized risk factor contribution map, which clearly showed that "TSH level" contributed a risk of +0.3, "sleep irregularity" contributed a risk of +0.15, and "regular exercise" contributed a protective effect of -0.05. This transformed the abstract risk score into an interpretable and intervention-guided attribution analysis.

[0028] S4: Perform time-series analysis on the calculated risk factors and risk scores to obtain a comprehensive risk score. Based on the comprehensive risk score, classify the risk level of each individual in the target population, mark the high-risk individuals, and generate a personalized screening strategy for each individual. The screening strategy includes a recommended combination of screening indicators and screening frequency.

[0029] In this embodiment, S4 performs time-series trend analysis on multiple risk scores within a preset historical time window, calculates the rate of risk change, and calculates a comprehensive risk score by combining the current risk score and risk factors through a weighted summation. The comprehensive risk score is compared with a preset risk level threshold range to classify each individual into low-risk, medium-risk, or high-risk levels, and individuals with high-risk levels are marked. For each individual, a personalized screening strategy is generated based on their risk level and the generated individualized risk factor contribution map. The generation of the screening strategy includes: When an individual is classified as high-risk, it is recommended to include a combination of screening indicators corresponding to the top N risk factors with the highest contribution, and set a first screening frequency. When an individual is classified as medium-risk, a standard screening indicator combination is recommended, and a second screening frequency lower than the first screening frequency is set. When an individual is classified as low-risk, a basic screening indicator combination is recommended, and a third screening frequency lower than the second screening frequency is set.

[0030] It should be specifically noted that this refers to the risk score sequence for a particular patient over the past 6 months. A time-series trend analysis was performed, calculating the risk change rate to be 0.07 / month, indicating a rapid upward trend in risk. Next, this change rate (weighted at 0.3) was weighted and summed with the current risk score of 0.75 (weighted at 0.7) to obtain a comprehensive risk score of 0.765. This score was then compared with a preset threshold (e.g., high risk). 0.7, medium risk 0.4-0.7, low risk 0.4) Comparison, classifying the patient as high-risk and marking them accordingly. Finally, a personalized screening strategy was generated for them: based on their individualized risk factor contribution profile, the top three contributing risk factors were identified as "TSH level fluctuations," "irregular sleep," and "high-iodine environment." Therefore, a screening indicator combination including "five thyroid function tests, sleep monitoring questionnaire, and urinary iodine test" was recommended, with a first screening frequency of once every 3 months; if the patient was of medium risk, a standard screening indicator combination including only "five thyroid function tests" was recommended, with a second screening frequency of once every 6 months; if the patient was of low risk, a basic screening indicator combination including only "TSH single test" was recommended, with a third screening frequency of once a year.

[0031] S5: Based on the high-risk individuals and their historical multi-source thyroid information data, analyze the key driving paths that lead to the continuous rise in their risk scores.

[0032] In this embodiment, step S5 constructs a directed acyclic graph containing the characteristic variables in the historical multi-source information data of the high-risk individual as a causal hypothesis graph. The nodes of the directed acyclic graph represent the characteristic variables, and the directed edges between the nodes represent the preset causal relationship. A causal effect estimation model based on the backdoor criterion is adopted to identify all backdoor paths between any two target variables on the causal hypothesis graph, and the causal effect value represented by the directed edge is calculated by adjusting the set of confounding variables. Traverse all complete paths from the root node with no incoming edges to the target node representing the occurrence of the risk score in the directed acyclic graph, and multiply the average causal effect values ​​of each directed edge on each path to obtain the total causal effect value of the path. Based on the calculated total causal effect value of each path, the paths are sorted in descending order, and the path with the highest total causal effect value is identified and output as the key driving path that causes the risk score of the high-risk individual to continue to rise.

[0033] It needs to be specifically explained that, firstly, a directed acyclic graph (DAG) is constructed as a causal hypothesis graph, containing characteristic variables such as "high-iodine diet," "mental stress," "insufficient sleep," "elevated TSH levels," and "hyperthyroidism risk score." This DAG presupposes that "high-iodine diet" and "mental stress" both point to "insufficient sleep," "insufficient sleep" points to "elevated TSH levels," and "elevated TSH levels" ultimately points to "hyperthyroidism risk score." Next, a causal effect estimation model based on the backdoor criterion is used. For example, when identifying the causal effect from "insufficient sleep" to "elevated TSH levels," a backdoor path is found: "insufficient sleep..." Mental stress "TSH levels increased," so by adjusting for the confounding variable "mental stress," the average causal effect value of this directed edge was calculated to be +0.4 (i.e., for every unit increase in sleep deprivation, TSH levels increased by an average of 0.4 units). Subsequently, the complete path from the root node "high-iodine diet" to the target node "hyperthyroidism risk score" was traversed. sleep deprivation SH levels increased The "hyperthyroidism risk score" is calculated by multiplying the average causal effect values ​​of the three directed edges along this path (assumed to be 0.8, 0.4, and 0.6 respectively) to obtain the total causal effect value of this path. Finally, the total causal effect values ​​of all pathways are sorted in descending order. If the total effect value of the pathway starting with "high iodine diet" is the highest, it is identified and output as the key driving pathway that leads to the patient's continued high risk.

[0034] S6: Perform correlation analysis between the analyzed key driving paths and the external knowledge base to predict the recurrence frequency of the key driving paths.

[0035] In this embodiment, step S6 performs semantic matching and mapping between the nodes and directed edges in the key driving path and the standardized medical ontology and causal associations pre-stored in the external knowledge base, so as to identify the standardized causal path corresponding to the key driving path in the external knowledge base. From the external knowledge base, retrieve multiple historical case data associated with the standardized causal path, and extract the recurrence event records of the corresponding path in each historical case; Based on the historical case data, a survival analysis model is constructed, using the path length, node type, and causal effect value of each directed edge of the key driving path as covariates, and the recurrence event of the path as the endpoint event, to calculate the recurrence probability of the key driving path within a future preset time window. The calculated recurrence probability is output as the predicted recurrence frequency value of the key driving path.

[0036] It should be specifically noted that the key driving pathway identified by S5 analysis, "high-iodine diet," sleep deprivation TSH levels increased The "hyperthyroidism risk score" was semantically matched with external knowledge bases (such as Systematized Nomenclature of Medicine-Clinical Terms, SNOMEDCT) and mapped to the standardized causal path "excessive iodine intake". Sleep disorders Abnormal thyroid-stimulating hormone levels "Hyperthyroidism". Next, from the tens of thousands of historical cases of hyperthyroidism patients associated with the knowledge base, all cases containing this standardized path were retrieved, and records of recurrence of path-related symptoms after treatment were extracted. Then, based on these historical case data, a Cox proportional hazards model was constructed as a survival analysis model, encoding the path length (4), node type (such as "behavior", "biochemistry", "disease", etc.), and the causal effect value of each directed edge calculated by S5. The model inputs these values ​​as covariates. Finally, the model calculates a 65% probability of recurrence of this key driving path within the next two years, which is then output as the predicted recurrence frequency for this path.

[0037] S7: Based on the recurrence frequency of key driving pathways, dynamically adjust and optimize the screening strategy and the intervention strategy corresponding to the key driving pathways.

[0038] In this embodiment, S7 compares the predicted recurrence frequency of the key driving path with a preset frequency threshold, and divides each key driving path into a high recurrence risk path or a low recurrence risk path. For the high recurrence risk pathway, the corresponding screening and intervention strategies are strengthened and adjusted. The strengthening and adjustment of the screening strategy includes: adding specific screening indicators for key nodes of the pathway and increasing the screening frequency, based on the original individualized screening strategy. The strengthening and adjustment of the intervention strategy includes: querying multiple intervention measures matching the node type of the high recurrence risk pathway from a preset intervention strategy knowledge base, and weighting and ranking the intervention measures according to their pathway suppression success rates recorded in historical cases to generate a combined intervention plan containing multiple intervention measures. For the aforementioned low recurrence risk pathways, maintain or reduce the corresponding screening frequency and adopt standardized intervention protocols based on clinical guidelines; By integrating all key driver-adjusted screening and intervention strategies, a personalized health management plan is generated. Finally, the generated personalized health management plan and recurrence frequency prediction value are output and sent to the target terminal via wireless communication and displayed in a visual interface. The target terminal includes, but is not limited to, doctor workstations and patient mobile terminal devices, for doctors and patients to view and execute.

[0039] It should be specifically noted that the predicted key driver pathway for a certain patient, "high-iodine diet," will be... sleep deprivation The recurrence frequency of "elevated TSH levels" (65%) was compared with the preset threshold of 50%, classifying it as a high-risk recurrence pathway. Subsequently, this pathway was strengthened and adjusted: In terms of screening strategy, in addition to the existing thyroid function tests every 3 months, specific screening indicators targeting the "sleep deprivation" node (such as polysomnography reports) were added, and the screening frequency was increased to every 2 months; in terms of intervention strategy, intervention measures matching the "behavioral" (high-iodine diet) and "physiological" (sleep deprivation) nodes were searched from the intervention knowledge base, such as "low-iodine diet guidance" and "cognitive behavioral therapy (CBT)." The study combined "melatonin supplementation" and other treatments with a weighted ranking based on their success rates in suppressing pathways in historical cases (70%, 85%, and 60%, respectively), generating a combined intervention plan that includes "low-iodine dietary guidance" and "cognitive behavioral therapy." Finally, this enhanced plan was integrated with maintenance strategies for other low-relapse-risk pathways to generate a complete individualized health management plan. This plan was then wirelessly transmitted to the doctor's workstation and the patient's mobile app, providing a visual interface that displays the high-relapse-risk pathway, the predicted relapse probability, and the corresponding enhanced intervention measures for joint confirmation and implementation by both the doctor and patient.

[0040] likeFigure 2 This embodiment provides an implementation system for a method of full-cycle thyroid health management, including a data acquisition module, a feature extraction module, a risk calculation module, a comprehensive risk assessment module, a key driving path analysis module, a recurrence prediction module, and an optimization feedback module. The data acquisition module is connected to the feature extraction module, the feature extraction module is connected to the risk calculation module, the risk calculation module is connected to the comprehensive risk assessment module, the comprehensive risk assessment module is connected to the key driving path analysis module, the key driving path analysis module is connected to the recurrence prediction module, and the comprehensive risk assessment module, the recurrence prediction module, and the optimization feedback module are all connected.

[0041] The data acquisition module acquires multi-source thyroid information data of the target population and forms a multi-source information dataset; The feature extraction module extracts and fuses features from the multi-source information dataset to generate a time-series feature vector characterizing an individual's thyroid health status. The risk calculation module analyzes and calculates the risk factors and risk scores for the occurrence of thyroid diseases for each feature dimension within a preset time window based on the generated time-series feature vector. The risk comprehensive assessment module performs time-series analysis on the calculated risk factors and risk scores to obtain a comprehensive risk score. Based on the comprehensive risk score, it classifies each individual in the target population into risk levels, marks high-risk individuals, and generates a personalized screening strategy for each individual. The screening strategy includes a recommended combination of screening indicators and screening frequency. The key driving path analysis module analyzes the key driving paths that cause the risk score to continue to rise based on the high-risk individuals and their historical multi-source thyroid information data. The recurrence prediction module performs correlation analysis between the analyzed key driving paths and an external knowledge base to predict the recurrence frequency of the key driving paths. The optimization feedback module dynamically adjusts and optimizes the screening strategy and the intervention strategy corresponding to the key driving path based on the recurrence frequency of the key driving path.

[0042] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0043] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for thyroid cycle health management, characterized by, The method comprises the following steps: S1: Obtain thyroid multi-source information data of a target population, and form a multi-source information data set; S2: Perform feature extraction and fusion on the multi-source information data set to generate a time sequence feature vector representing the health status of an individual thyroid; S3: According to the generated time sequence feature vector, analyze and calculate the risk factors and risk scores of each feature dimension for the occurrence of thyroid disease within a preset time window; S4: Perform time sequence analysis on the calculated risk factors and risk scores to obtain a comprehensive risk score, and based on the comprehensive risk score, divide each individual in the target population into a risk level, mark the individuals in the high risk level, and generate a personalized screening strategy for each individual, which includes recommended screening index combination and screening frequency; S5: According to the individuals in the high risk level, analyze the key driving path that leads to the continuous increase of the risk score of the high risk individuals based on the historical thyroid multi-source information data of the high risk individuals; S6: Correlate the analyzed key driving path with an external knowledge base to predict the recurrence frequency of the key driving path; S7: According to the recurrence frequency of the key driving path, dynamically adjust and optimize the screening strategy and the intervention strategy corresponding to the key driving path.

2. The method for thyroid cycle health management according to claim 1, wherein, The S1 obtains multi-source information data of thyroid patients from multiple data sources through a multi-modal perception terminal, wherein the multi-source information data includes clinical data, behavior data, environmental data, and psychological data; A cross-modal feature alignment network is used to map the multi-source heterogeneous data to a unified semantic space to generate a high-dimensional multi-source information data set.

3. The method for thyroid cycle health management according to claim 2, wherein, The S2 includes the following steps of feature extraction and fusion on the multi-source information data set: The thyroid function index imaging features and pathological section information in the clinical data are processed by a deep time sequence convolution network to extract the dynamic pattern in the clinical data over time, and a first time sequence feature vector is generated; wherein the deep time sequence convolution network is stacked by at least two residual convolution blocks, each residual convolution block includes a one-dimensional convolution layer, an activation function layer and a dilated convolution layer, which are used to capture the evolution dynamic pattern of different time scales in the clinical data; The medication adherence, exercise mode in the behavior data, and regional iodine exposure level in the environmental data are used to construct an association graph, and a graph neural network is used to process the association graph to mine the non-linear association between the behavior data and the environmental data, and a graph structure feature vector is generated; wherein the construction of the association graph includes: taking medication adherence, exercise mode and regional iodine exposure level as nodes of the graph, and determining the weight of the edges between the nodes according to the preset medical rules or the Pearson correlation coefficient between the data; The emotion fluctuation and stress index in the psychological data are processed by a long short-term memory network to capture the long time sequence dependence in the psychological data, and a second time sequence feature vector is generated; The extracted first time sequence feature vector, the graph structure feature vector and the second time sequence feature vector are fused by a cross-modal attention fusion mechanism to generate a time sequence feature vector representing the health status of the individual thyroid.

4. The method for thyroid cycle health management according to claim 3, wherein, The S3 assigns weights to each dimension of the time sequence feature vector within a preset time window through a time sequence attention network to quantify the differential effects of clinical indicator fluctuations, behavior pattern changes, environmental exposure accumulations and psychological state evolutions on the thyroid health status at different time points, thereby dynamically identifying key risk factors. The weighted feature vector is input into a risk prediction model trained on a disease-specific dataset containing thyroid function indicators, imaging features and outcome labels, which is used to calculate the risk score of thyroid disease occurrence at a specific future time point. A model explanation method based on game theory is used to reverse analyze the decision-making process of the risk prediction model, decompose the risk score into the contribution of each feature dimension, and generate an individualized risk factor contribution map.

5. The method for thyroid cycle health management according to claim 4, wherein, The S4 performs time sequence trend analysis on multiple occurrence risk scores within a preset historical time window, calculates the risk change rate, and combines the current occurrence risk score and occurrence risk factors to calculate a comprehensive risk score by weighted summation. The comprehensive risk score is compared with a preset risk level threshold interval to divide each individual into a low risk, medium risk or high risk level, and mark the high risk level individuals. For each individual, a personalized screening strategy is generated based on its risk level and the generated individualized risk factor contribution map, which includes: When the individual is classified as a high risk level, the top N risk factors corresponding to the screening indicator combination with the highest contribution are recommended, and a first screening frequency is set. When the individual is classified as a medium risk level, a standard screening indicator combination is recommended, and a second screening frequency lower than the first screening frequency is set. When the individual is classified as a low risk level, a basic screening indicator combination is recommended, and a third screening frequency lower than the second screening frequency is set.

6. The method for thyroid cycle health management according to claim 5, wherein, The S5 constructs a directed acyclic graph including each feature variable in the individual's historical thyroid multi-source information data as a causal hypothesis graph for the high risk level individuals, wherein the nodes of the directed acyclic graph represent the feature variables, and the directed edges between the nodes represent preset causal relationships. A causal effect estimation model based on the backdoor criterion is used to identify all backdoor paths between any two target variables on the causal hypothesis graph, and by adjusting the set of confounding variables, the causal effect value represented by the directed edge is calculated. All complete paths from the root node with no incoming edge to the target node representing the occurrence risk score in the directed acyclic graph are traversed, the average causal effect value of each directed edge on each path is multiplied, and the total causal effect value of the path is obtained. According to the total causal effect value of each path calculated, the paths are sorted in descending order, and the path with the highest total causal effect value is identified and output as the key driving path that causes the individual risk score of the high risk level to continue to rise.

7. The method for thyroid cycle health management according to claim 6, wherein, The S6 performs semantic matching and mapping of the nodes and directed edges in the key driving path with the pre-stored standardized medical ontology and causal association in the external knowledge base to identify the corresponding standardized causal path of the key driving path in the external knowledge base. From the external knowledge base, a plurality of historical case data associated with the standardized causal path are retrieved, and the recurrence event records of the corresponding path in each historical case are extracted. Based on the historical case data, a survival analysis model is constructed, the path length, node type, and causal effect value of each directed edge of the key driving path are taken as covariates, and the recurrence event of the path is taken as the endpoint event, and the recurrence probability of the key driving path within a preset time window in the future is calculated. The calculated recurrence probability is output as the recurrence frequency prediction value of the key driving path.

8. The method for thyroid cycle health management according to claim 7, wherein, The S7 compares the recurrence frequency prediction value of the key driving path with a preset frequency threshold, and divides each key driving path into a high recurrence risk path or a low recurrence risk path. For the high recurrence risk path, its corresponding screening strategy and intervention strategy are adjusted; wherein the adjustment of the screening strategy includes: on the basis of the original individualized screening strategy, adding special screening indicators for the key nodes of the path and increasing the screening frequency; the adjustment of the intervention strategy includes: querying a plurality of intervention measures matching the node type of the high recurrence risk path from a preset intervention strategy knowledge base, and generating a combined intervention scheme including a plurality of intervention measures according to the path suppression success rate recorded in the historical cases. For the low recurrence risk path, the screening frequency is maintained or reduced, and a standardized intervention scheme based on clinical guidelines is adopted. Integrate the adjusted screening strategies and intervention strategies of all key driving paths to generate an individualized health management plan, and finally output and send the generated individualized health management plan and recurrence frequency prediction value to the target terminal through wireless communication and display in a visual interface, the target terminal includes but is not limited to a doctor workstation, a patient mobile terminal device, for doctors and patients to view and execute.

Images