Information Management System and Method for Stroke Patients in the Department of Neurology

Abstract

The present invention discloses an information-based management system and method for stroke patients in the department of neurology, comprising the following steps: constructing a standardized multi-modal data set by collecting data such as images, physiological signals, genetic information, and electronic medical records; identifying pathological features and their causal relationships by using causal inference methods, generating a disease causal diagram, and eliminating confounding factors to optimize the accuracy of data analysis; extracting shared features of multi-modal data based on cross-modal representation learning methods to establish a personalized health profile; evaluating the impact of different intervention measures on the development of the disease by combining time-series causal inference methods, and dynamically adjusting the rehabilitation path by using reinforcement learning methods; calculating the optimal rehabilitation path by causal dynamic programming methods, and collecting feedback data in real time during the rehabilitation process to optimize the intervention strategy. The present invention can improve the intelligent and precise level of stroke patient management, realize personalized intervention and rehabilitation path optimization, and improve the rehabilitation effect of patients.

Description

Technical Field

[0001] The present invention relates to the technical field of information management, and particularly to an information management system and method for stroke patients in the department of neurology. Background Art

[0002] With the rapid development of medical informatization, the management methods for stroke patients in the department of neurology are also evolving continuously. However, there are still many technical bottlenecks in the existing stroke patient management systems in terms of information integration, personalized intervention, and rehabilitation path optimization. The development of the condition of stroke patients involves multi-modal data, and the heterogeneity and information isolation between different data sources seriously affect the effective utilization of data. In addition, the existing management methods lack accurate condition analysis and personalized rehabilitation intervention strategies, resulting in difficult optimization of the rehabilitation effect of patients.

[0003] In the prior art, traditional stroke management systems usually rely on rule-based decision-making models based on a single data source or rehabilitation programs guided by experience. These methods are often difficult to provide efficient and accurate management programs when facing complex and changeable disease progressions. Specifically, the prior art mainly has the following defects:

[0004] 1. Data islands and incoherence: The existing stroke management systems cannot effectively integrate multi-modal data such as images, physiological signals, and genetic information, resulting in isolated data and incoherent information, which affects the accuracy of condition analysis.

[0005] 2. Lack of causal reasoning analysis: Traditional condition assessment methods mostly rely on statistical models, are difficult to eliminate confounding factors, and cannot accurately model the disease evolution process, which affects the scientific nature of personalized intervention decisions.

[0006] 3. Difficulty in cross-modal data alignment: The existing methods lack an effective cross-modal learning mechanism and are difficult to extract shared features from different data sources, resulting in inaccurate individual health portraits and unable to provide accurate personalized intervention programs.

[0007] 4. Insufficient optimization of intervention measures: Traditional stroke management mostly adopts static rehabilitation path planning and fails to make full use of intelligent optimization technologies such as reinforcement learning, resulting in the lack of dynamic adjustment ability of rehabilitation intervention strategies and being difficult to adapt to the real-time changes of patients' conditions.

[0008] 5. Lack of long-term optimization of the rehabilitation path: The existing rehabilitation programs do not fully consider the long-term effects of different intervention measures and lack causal dynamic programming methods to optimize the rehabilitation path, which may lead to an extended rehabilitation period or unsatisfactory recovery effect of patients.

[0009] Therefore, how to provide an information management system and method for stroke patients in the department of neurology is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0010] An object of the present invention is to propose an information management system and method for stroke patients in the department of neurology. The present invention makes full use of multi-modal data fusion, temporal causal reasoning, and dynamic rehabilitation path optimization technologies, and details the methods for modeling the disease evolution of stroke patients, optimizing personalized interventions, and dynamically adjusting rehabilitation paths, with the advantages of high data integration, high intervention accuracy, and intelligent rehabilitation optimization.

[0011] The information management system and method for stroke patients in the department of neurology according to the embodiments of the present invention include the following steps:

[0012] S1. A data acquisition module, which acquires image data, physiological signals, genetic information, and electronic medical record data of stroke patients in the department of neurology, performs standardized processing and storage, removes redundant information, and constructs a structured multi-modal data set;

[0013] S2. A causal reasoning analysis module, which, based on the structured multi-modal data set, uses a structural causal learning method to analyze pathological features, eliminates confounding factors, constructs a causal model of disease evolution, and generates a disease causal diagram;

[0014] S3. A cross-modal learning module, which, based on the disease causal diagram, uses a self-supervised cross-modal representation learning method to extract shared features of the structured multi-modal data set, performs feature alignment and multi-modal mapping, and generates a patient health portrait;

[0015] S4. A disease evolution analysis module, which, based on the patient health portrait and the disease causal diagram, combines temporal causal reasoning to evaluate the impact of different intervention measures on the development of the disease and forms an intervention plan;

[0016] S5. A data fusion module, which performs cross-modal consistency verification on the disease causal diagram, the patient health portrait, and the intervention plan, and uses an adaptive weighting strategy to adjust the data contribution degree to generate rehabilitation path decision-making data;

[0017] S6. A rehabilitation optimization decision-making module, which, based on the rehabilitation path decision-making data, uses a causal dynamic programming method to simulate the long-term impact of different intervention plans and dynamically adjusts the intervention plan during the execution of the rehabilitation path.

[0018] Optionally, the S2 specifically includes:

[0019] S21. A data parsing unit, which receives the structured multi-modal data set, including image data, physiological signals, genetic information, and electronic medical record data, performs modal separation on the structured multi-modal data set, extracts time series features and static features, and generates a standardized pathological data matrix;

[0020] S22. Causality Modeling Unit: Based on the standardized pathological data matrix, a causal model of disease progression is constructed using structural causal learning methods to identify causal relationships between variables and generate a causal association matrix. , where represents the causal influence weight of variable on . Variables and variable respectively represent variables related to the disease condition.

[0021] S23. Confounding Factor Elimination Unit: Based on the causal association matrix, a causal intervention inference method is used to calculate the direct causal effects between different variables to eliminate confounding factors and generate an adjusted causal association matrix:

[0022] ;

[0023] where is the causal effect of variable on the disease state after eliminating the influence of confounding factor under the constraint of the causal association matrix . is the probability of the disease state under the combined action of variables, confounding factors, and the causal association matrix. is the prior probability distribution of confounding factor under the constraint of the causal association matrix .

[0024] S24. Disease Progression Modeling Unit: Based on the adjusted causal association matrix , combined with time series analysis methods, calculate the change trend of the disease state and establish a time-series causal relationship network to evaluate the disease development path.

[0025] S25. Causal Graph Generation Unit: Based on the causal model of disease progression, construct a disease causal graph , where is the variable set, is the causal association relationship, and the causal association relationship is the time-dependent causal relationship of the time-series causal relationship network, and the disease causal graph is transmitted to the cross-modal learning module.

[0026] Optionally, the S23 specifically includes:

[0027] S231. Confounding Factor Identification Unit: Based on the causal association matrix , identify the confounding factor , and use the conditional independence test method to evaluate variable and the disease state Independence between, by calculating the conditional independence probability Determine confounding factors Whether it has an impact on the variable and the disease status in the causal relationship between;

[0028] S232, Causal effect calculation unit, based on the causal association matrix and confounding factors , using the causal intervention inference method to calculate the variable 's direct causal effect on the disease status , eliminating the influence of confounding factors and generating the causal effect calculation result after eliminating confounding factors;

[0029] S233, Adjustment matrix generation unit, based on the causal effect calculation result after eliminating confounding factors, generate an adjusted causal association matrix , where represents eliminating confounding factors and then optimizing the calculation of the adjusted causal association matrix using the Bayesian update method.

[0030] Optionally, the S3 specifically includes:

[0031] S31, Cross-modal feature extraction unit, receive the disease causal graph , based on the structured multi-modal dataset, construct a standardized feature matrix ;

[0032] S32, Use the self-supervised cross-modal representation learning method to construct a cross-modal feature mapping model , and perform feature encoding on the standardized feature matrix to extract the latent shared feature representation :

[0033] ;

[0034] Among them, is the latent shared feature representation, is the cross-modal feature mapping model, is the disease causal graph;

[0035] S33, Feature alignment unit, based on the latent shared feature representation , calculate the feature similarity between different modal data, optimize using the cross-modal contrast learning loss function, minimize the feature distribution distance of different modal features of the same patient in the latent shared feature representation, maximize the feature distribution distance of different patients, and generate the optimized shared feature representation ;

[0036] S34. The healthy image generation unit generates a patient's healthy image based on the optimized shared feature representation , adopting a weighted feature fusion strategy :

[0037] ;

[0038] Among them, is the feature fusion weight, is the fusion bias term, is the patient's healthy image

[0039] Optionally, the S4 specifically includes:

[0040] S41. The disease state modeling unit extracts an individual pathological feature vector based on the patient's healthy image , constructs a disease feature space, combines a temporal causal reasoning method, and calculates the causal association weight between different pathological features to quantify the evolution process of the disease at consecutive time steps;

[0041] S42. The intervention plan generation unit constructs a set of intervention measures based on the disease feature space where is an intervention measure, is the type of intervention measure, defines an intervention measure influence function calculates the influence value of different intervention measures on the individual pathological feature vector and constructs an intervention measure evaluation matrix :

[0042] ;

[0043] Among them, is used to measure the contribution degree of the intervention measure to the individual pathological feature vector ;

[0044] S43. The reinforcement learning optimization unit optimizes the intervention measures based on the intervention measure evaluation matrix , adopts a reinforcement learning method to optimize the intervention measures, calculates the long-term return value of the intervention measures, and selects the optimal disease intervention path through a decision optimization function :

[0045] ;

[0046] Among them, represents the long-term return value obtained by applying the intervention measure under the individual pathological feature vector , The time window for the evaluation of the intervention measure;

[0047] S44. An intervention path update unit, based on the individual pathological feature vector and the long-term return value of the intervention measure, dynamically adjusts the intervention measure to form an intervention plan.

[0048] Optionally, the S6 specifically includes:

[0049] S61. A rehabilitation status modeling unit, which receives the rehabilitation path decision data and constructs a rehabilitation status set , where represents the rehabilitation status of the patient at time ;

[0050] S62. A causal dynamic programming unit, based on the rehabilitation status set and the intervention measure set , uses the causal dynamic programming method to simulate the influence of different rehabilitation paths on the long-term recovery effect, defines the optimal value function of the rehabilitation path , and calculates the optimal value of the rehabilitation path for each rehabilitation status :

[0051] ;

[0052] Among them, represents the immediate rehabilitation benefit obtained by performing the intervention measure in the rehabilitation status , is the discount factor, which measures the balance between the current rehabilitation benefit and the long-term rehabilitation impact, is the next moment's rehabilitation status of the patient after the intervention, represents the optimal value of the rehabilitation path of the rehabilitation status ;

[0053] S63. A rehabilitation path optimization unit, based on the optimal values of the rehabilitation paths of each rehabilitation status, calculates the optimal rehabilitation path;

[0054] S64. A personalized adjustment unit, which uses an adaptive update mechanism to adjust the optimal rehabilitation path, evaluates the patient's recovery progress through dynamic causal reasoning, and optimizes the intervention plan to make the intervention plan adapt to the individual recovery needs.

[0055] Optionally, for the information-based management method of stroke patients in neurology department, the method includes the following steps:

[0056] S71. Collect the image data, physiological signals, gene information and electronic medical record data of stroke patients in the neurology department, perform data format conversion and normalization processing, remove redundant and abnormal data, and construct a structured multimodal dataset;

[0057] S72. Based on the structured multi-modal dataset, construct a causal model of disease progression using structural causal learning method, identify variables and their causal associations, generate a disease causal graph, and eliminate confounding factors to optimize the causal association matrix;

[0058] S73. Based on the disease causal graph, use self-supervised cross-modal representation learning method to extract shared features of the structured multi-modal dataset, perform cross-modal feature alignment and multi-modal mapping to generate a patient health portrait;

[0059] S74. Based on the patient health portrait and the disease causal graph, combine the temporal causal reasoning method to predict the impact of different intervention measures on the disease development, and use the reinforcement learning method to optimize the disease intervention measures to form an intervention plan;

[0060] S75. Conduct cross-modal consistency verification on the disease causal graph, patient health portrait and intervention plan, adopt an adaptive weighted strategy to adjust the data contribution degree, and generate rehabilitation path decision-making data;

[0061] S76. Based on the rehabilitation path decision-making data, combine the causal dynamic programming method to simulate the long-term impact of different intervention plans, calculate the optimal rehabilitation path, and dynamically adjust the intervention plan during the execution of the rehabilitation path.

[0062] The beneficial effects of the present invention are as follows:

[0063] (1) By combining causal reasoning, cross-modal learning and reinforcement learning optimization, the present invention provides a deep understanding and dynamic adaptation of the disease development of stroke patients, enabling the system to construct an accurate disease causal model based on multi-modal data, identify the causal relationships between pathological features, and eliminate confounding factors, thereby improving the accuracy of disease assessment. Using the cross-modal representation learning method, it can efficiently perform feature alignment and fusion on multi-modal data such as images, physiological signals, gene data and electronic medical records to form an accurate individualized health portrait to support personalized intervention decisions.

[0064] (2) Through temporal causal reasoning and reinforcement learning optimization, combining the individual health portrait and the disease causal graph, the present invention can evaluate the impact of different intervention measures on the disease development in real time and optimize the personalized intervention path. This not only improves the accuracy of stroke intervention strategies, but also can adaptively adjust the intervention plan during the dynamic evolution of the patient's condition to ensure the selection of the optimal rehabilitation path and improve the long-term rehabilitation effect of the patient.

[0065] (3) The present invention simulates the long-term effects of different rehabilitation intervention programs through causal dynamic programming, constructs an optimized rehabilitation path decision-making model, and dynamically updates the intervention strategy during the rehabilitation execution process. The reinforcement learning is used to optimize the rehabilitation path, enabling the system to make intelligent decisions based on the real-time condition changes of patients, ensuring the continuous optimization of the rehabilitation program, improving the rehabilitation quality and management efficiency of stroke patients. This method reduces the limitations of traditional rehabilitation path relying on empirical decision-making and realizes a personalized, intelligent and dynamically optimized stroke management program. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0067] Figure 1 FIG. is a system architecture diagram of the information management system for stroke patients in the department of neurology proposed by the present invention;

[0068] Figure 2 FIG. is a flowchart of the information management method for stroke patients in the department of neurology proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic way, so they only show the components related to the present invention.

[0070] Refer to Figure 1-2 , the information management system and method for stroke patients in the department of neurology include the following steps:

[0071] S1. A data acquisition module collects the image data, physiological signals, gene information and electronic medical record data of stroke patients in the department of neurology, performs standardized processing and storage, removes redundant information, and constructs a structured multi-modal data set;

[0072] In this embodiment, image data, physiological signals, genetic information, and electronic medical record data of stroke patients in the department of neurology are collected. Among them, the image data includes MRI, CT, PET-CT, etc., the physiological signals include blood pressure, electrocardiogram, electroencephalogram, blood oxygen saturation, etc., the genetic information includes genes such as APOE, CYP2C19, MTHFR, etc., and the electronic medical record data integrates the patient's basic medical history, medication records, laboratory tests, neurological function scores, and rehabilitation training conditions, etc. The data standardization processing technology is used to remove redundant information, ensure the consistency of the data formats of different data sources, and construct a structured multi-modal data set to support subsequent causal inference analysis and the generation of personalized health portraits. This method can effectively integrate heterogeneous data, improve the data quality and integrity, provide a reliable data basis for the condition assessment and precise intervention of stroke patients, and achieve more precise condition monitoring and dynamic management.

[0073] S2. Causal inference analysis module, based on the structured multi-modal data set, uses the structural causal learning method to analyze pathological features, eliminate confounding factors, construct a causal model of disease evolution, and generate a causal diagram of the disease;

[0074] S3. Cross-modal learning module, based on the causal diagram of the disease, uses the self-supervised cross-modal representation learning method to extract the shared features of the structured multi-modal data set, perform feature alignment and multi-modal mapping, and generate a patient health portrait;

[0075] S4. Disease evolution analysis module, based on the patient health portrait and the causal diagram of the disease, combines temporal causal inference to evaluate the impact of different intervention measures on the disease development, and forms an intervention plan;

[0076] S5. Data fusion module, performs cross-modal consistency verification on the causal diagram of the disease, the patient health portrait, and the intervention plan, uses an adaptive weighted strategy to adjust the data contribution degree, and generates rehabilitation path decision data;

[0077] This embodiment adopts a cross-modal consistency verification mechanism to ensure the consistency of the causal diagram of the disease, the health portrait, and the intervention plan during the multi-modal data fusion process. Based on the statistical consistency test method, the distribution similarity between different modal data is calculated. The maximum mean difference is used to measure the cross-modal distribution deviation of image data, physiological signals, genetic information, and electronic medical record data. The cross-modal feature alignment is optimized through an adversarial training strategy. The cross-modal contrast learning method is used to construct a feature mapping network, and the feature mutual information between different data sources is calculated to ensure the consistency of the representations of different modal data in the shared space. The adaptive weighted fusion strategy is adopted to optimize the weight allocation of each modal data, reduce the impact of abnormal data on the overall model, improve the reliability of disease assessment and rehabilitation path decision-making, and thus improve the accuracy and effectiveness of personalized rehabilitation plans.

[0078] S6. Rehabilitation Optimization Decision-making Module. Based on the rehabilitation path decision-making data, it uses the causal dynamic programming method to simulate the long-term impacts of different intervention plans and dynamically adjusts the intervention plan during the execution of the rehabilitation path.

[0079] In this embodiment, the said S2 specifically includes:

[0080] S21. Data Analysis Unit. It receives the structured multi-modal data set, including image data, physiological signals, gene information, and electronic medical record data, performs modal separation on the structured multi-modal data set, extracts time series features and static features, and generates a standardized pathological data matrix;

[0081] S22. Causal Relationship Modeling Unit. Based on the standardized pathological data matrix, it uses the structural causal learning method to construct a causal model of the disease evolution, identifies the causal relationships between variables, and generates a causal association matrix , where represents the causal influence weight of variable on , and variables and variable respectively represent the variables related to the disease condition;

[0082] S23. Confounding Factor Elimination Unit. Based on the causal association matrix, it uses the causal intervention reasoning method to calculate the direct causal effects between different variables to eliminate the confounding factors and generate an adjusted causal association matrix:

[0083] ;

[0084] where, is the causal effect of variable on the disease condition after eliminating the influence of the confounding factor under the constraint of the causal association matrix , is the probability of the disease condition under the combined action of variables, confounding factors, and the causal association matrix, is the prior probability distribution of the confounding factor under the constraint of the causal association matrix ;

[0085] This formula is used to calculate the change in the disease state under specific intervention measures and eliminate the influence of confounding factors on the causal relationship. Its principle is based on the intervention calculation method in causal inference. Through summation operations, the influence of confounding factors is marginalized, so that the final calculation result can truly reflect the direct impact of the intervention measures on the disease state. Specifically, this formula first calculates the probability of the disease state under the combined action of the intervention variable and the confounding factors, and then combines the prior probability of the confounding factors to finally obtain the probability distribution of the disease state after eliminating the influence of the confounding factors, thereby optimizing the accuracy of disease prediction and improving the scientific nature of intervention decisions.

[0086] S24. Disease evolution modeling unit, based on the adjusted causal association matrix , combined with time series analysis methods to calculate the disease state 's change trend and establish a time-series causal relationship network to evaluate the disease development path;

[0087] S25. Causal graph generation unit, based on the disease evolution causal model, constructs a disease causal graph , where is a variable set, is the causal association relationship, and the causal association relationship is the time-dependent causal relationship of the time-series causal relationship network, and transmits the disease causal graph to the cross-modal learning module.

[0088] This embodiment effectively solves the problems of strong data heterogeneity, inaccurate disease analysis, and uneliminated influence of confounding factors in the prior art, improves the accuracy of disease modeling, and provides a solid data basis and causal inference support for personalized intervention decisions.

[0089] In this embodiment, the S23 specifically includes:

[0090] S231. Confounding factor identification unit, based on the causal association matrix , identifies confounding factors , and uses the conditional independence test method to evaluate the independence between the variable and the disease state . By calculating the conditional independence probability , it is determined whether the confounding factor affects the causal relationship between the variable and the disease state ;

[0091] S232. Causal effect calculation unit, based on the causal association matrix and the confounding factor , uses the causal intervention inference method to calculate the direct causal effect of the variable on the disease state , and eliminates the confounding factor Generate the calculation result of the causal effect after removing confounding factors;

[0092] S233. An adjustment matrix generation unit generates an adjusted causal association matrix based on the calculation result of the causal effect after removing confounding factors , where represents removing confounding factors After that, the Bayesian update method is used to optimize the adjusted causal association matrix Calculation.

[0093] In this embodiment, confounding factors are identified based on the causal association matrix, and the independence between variables and disease states is evaluated through the conditional independence test method, so as to determine the influence of confounding factors on the causal relationship. The direct causal effect of variables on disease states is calculated using the causal intervention reasoning method, and the interference of confounding factors on causal reasoning is removed to ensure the accuracy of causal reasoning. The Bayesian update method is used to optimize the causal association matrix to generate an adjusted causal association matrix, so as to improve the stability and reliability of causal reasoning. This embodiment effectively improves the accuracy of disease evolution analysis, ensures the scientificity of intervention decisions, avoids the causal reasoning bias caused by the influence of confounding factors in traditional statistical methods, and provides a more reliable theoretical basis for the personalized disease assessment and precise intervention of stroke patients.

[0094] In this embodiment, S3 specifically includes:

[0095] S31. A cross-modal feature extraction unit receives the disease causal graph and constructs a standardized feature matrix based on the structured multi-modal data set ;

[0096] S32. Use the self-supervised cross-modal representation learning method to construct a cross-modal feature mapping model and perform feature encoding on the standardized feature matrix to extract the latent shared feature representation :

[0097] ;

[0098] Among them, is the latent shared feature representation, is the cross-modal feature mapping model, is the disease causal graph;

[0099] S33. A feature alignment unit, based on the latent shared feature representation , calculate the feature similarity between different modality data, and optimize it using a cross-modal contrastive learning loss function, so that the feature distribution distance of different modality features of the same patient in the latent shared feature representation is minimized, and the feature distribution distance of different patients is maximized, and generate an optimized shared feature representation ;

[0100] In this embodiment, based on the latent shared feature representation, the feature similarity between different modality data (such as images, physiological signals, genetic data, and electronic medical records) is calculated, and it is optimized using a cross-modal contrastive learning loss function, so that the distribution distance of multi-modal features of the same patient in the shared feature space is minimized, while the feature distribution distance of different patients is maximized, thereby improving the accuracy of feature alignment. This method ensures that cross-modal data can be effectively fused in a unified feature space, reduces information loss and feature deviation, improves the accuracy of individual health portraits, provides more accurate data support for subsequent disease assessment, intervention optimization, and rehabilitation path decision-making, and thus enhances the intelligent and personalized capabilities of the system for stroke patient management.

[0101] S34. A health portrait generation unit, based on the optimized shared feature representation , generate a patient's health portrait using a weighted feature fusion strategy :

[0102] ;

[0103] Among them, is the feature fusion weight, is the fusion bias term, is the patient's health portrait.

[0104] In this embodiment, through cross-modal feature extraction, feature alignment, and weighted fusion strategies, the construction of a personalized health portrait for stroke patients is realized. A standardized feature matrix is constructed based on the disease causal graph and multi-modal data, and a self-supervised cross-modal representation learning method is used to extract the shared feature representation, thereby enhancing the information interoperability between different modality data. The cross-modal contrastive learning loss function is used to optimize feature alignment, so that the feature distributions of multi-modal data of the same patient in the shared feature space are as close as possible, while the data features of different patients are kept distinct, improving the accuracy of patient representation. Through the weighted feature fusion strategy, combined with the disease causal graph, personalized key pathological features are extracted to generate an accurate health portrait. This embodiment can effectively integrate and align heterogeneous medical data, ensure the integrity and consistency of the health portrait, improve the accuracy of disease management, and thus optimize personalized stroke intervention strategies and rehabilitation paths.

[0105] In this embodiment, the specific content of S4 includes:

[0106] S41. Disease state modeling unit, based on the patient's health profile Extract the individual pathological feature vector , construct a disease feature space, combine with the time-series causal reasoning method, calculate the causal association weights between different pathological features to quantify the evolution process of the disease at continuous time steps;

[0107] In this embodiment, the individual pathological feature vector is extracted based on the patient's health profile, a disease feature space is constructed, and combined with the time-series causal reasoning method, the causal association weights between different pathological features are calculated to quantify the dynamic influence between disease variables. By constructing a disease feature space, this method can accurately depict the change trend of the disease state at different time steps, thereby identifying the evolution pattern of key pathological features and predicting the disease development trajectory. This method improves the accuracy of disease assessment, enabling disease modeling not to be limited to static data analysis, but to be able to dynamically adjust the relevance of the disease state, providing a more scientific basis for personalized intervention and rehabilitation path optimization.

[0108] S42. Intervention plan generation unit, based on the disease feature space, construct a set of intervention measures , where is the intervention measure, is the type of intervention measure, define the intervention measure influence function Calculate the influence values of different intervention measures on the individual pathological feature vector and construct an intervention measure evaluation matrix :

[0109] ;

[0110] Among them, is used to measure the contribution degree of the intervention measure to the individual pathological feature vector ;

[0111] S43. Reinforcement learning optimization unit, based on the intervention measure evaluation matrix , use the reinforcement learning method to optimize the intervention measures, calculate the long-term return value of the intervention measures, and select the optimal disease intervention path through the decision optimization function :

[0112] ;

[0113] Among them, represents the long-term return value obtained by applying the intervention measure under the individual pathological feature vector , is the time window for intervention measure evaluation;

[0114] This formula is used to calculate the optimal selection of different rehabilitation intervention measures for the long-term rehabilitation path of patients. By using the reinforcement learning method to optimize the intervention strategy, the system can dynamically adjust the rehabilitation path at multiple time steps. Its principle is based on the idea of value iteration, calculating the cumulative rehabilitation rewards obtained by performing different intervention measures in a specific rehabilitation state, and introducing a discount factor to balance short-term and long-term rehabilitation benefits, so as to ensure that the system preferentially selects the path that can bring the greatest long-term rehabilitation effect. This method can effectively improve the adaptability of the rehabilitation strategy, optimize and adjust the rehabilitation path according to the dynamic changes of the patient's condition, and thus enhance the personalized rehabilitation effect.

[0115] S44. An intervention path update unit, based on the individual pathological feature vector and the long-term return value of the intervention measure, dynamically adjusts the intervention measure to form an intervention plan.

[0116] This embodiment constructs a disease condition feature space based on the patient's health portrait, and combines the time-series causal reasoning method to quantify the evolution process of the disease condition state, so as to accurately predict the disease development trend. The system evaluates the impact of different intervention measures on individual pathological features through the intervention measure impact function, and constructs an intervention measure evaluation matrix to provide a quantitative basis for the personalized intervention plan. The reinforcement learning method is used to optimize the intervention decision, calculate the long-term return values of different intervention paths, and select the optimal intervention path through the decision optimization function to ensure that the intervention measure can effectively promote the patient's rehabilitation. The system dynamically adjusts the intervention path based on the patient's real-time pathological feature vector and the long-term return value of the intervention measure to form an individualized rehabilitation plan. This embodiment not only improves the accuracy and adaptability of stroke patient management by accurately modeling the disease condition changes, optimizing the decision-making process of the intervention measure, and realizing personalized dynamic adjustment through reinforcement learning, but also can optimize the rehabilitation path to make the patient's rehabilitation process more efficient and scientific.

[0117] In this embodiment, the S6 specifically includes:

[0118] S61. A rehabilitation state modeling unit, which receives the rehabilitation path decision data and constructs a rehabilitation state set , where represents the rehabilitation state of the patient at time ;

[0119] S62. A causal dynamic programming unit, based on the rehabilitation state set and the intervention measure set , uses the causal dynamic programming method to simulate the impact of different rehabilitation paths on the long-term recovery effect, defines the optimal value function of the rehabilitation path , and calculates the optimal value of the rehabilitation path for each rehabilitation state :

[0120] ;

[0121] Wherein, represents the immediate rehabilitation benefit obtained by performing an intervention in the rehabilitation state under the condition and is the immediate rehabilitation benefit obtained by performing an intervention measure is the discount factor, which measures the balance between the current rehabilitation benefit and the long-term rehabilitation impact is the next moment's rehabilitation state of the patient after the intervention represents the optimal value of the rehabilitation path in the rehabilitation state ;

[0122] This formula is used to calculate the optimal value of the rehabilitation path, measure the impact of different rehabilitation intervention measures in the long-term recovery process, and its principle is based on causal dynamic programming. By accumulating the immediate rehabilitation benefits and considering the long-term benefits of future rehabilitation states, it balances the short-term intervention effect and the long-term rehabilitation goal. The discount factor is used to adjust the weight of the current benefit and the future benefit to ensure that both the current rehabilitation progress and the long-term rehabilitation path can be reasonably planned during the optimization process. This calculation method enables the rehabilitation plan to dynamically adapt to the evolution trend of the patient's condition, optimize the selection of intervention measures, and thus improve the accuracy and long-term effectiveness of the personalized rehabilitation path.

[0123] S63. A rehabilitation path optimization unit, which calculates the optimal rehabilitation path based on the optimal values of the rehabilitation paths for each rehabilitation state;

[0124] S64. A personalized adjustment unit, which adjusts the optimal rehabilitation path by using an adaptive update mechanism, evaluates the patient's recovery progress through dynamic causal reasoning, optimizes the intervention plan, and makes the intervention plan adapt to the individual recovery needs.

[0125] In this embodiment, by constructing a set of rehabilitation states and a state transition matrix, the key state changes in the patient's rehabilitation process are quantified to ensure the scientificity and continuity of the rehabilitation path optimization. Based on the causal reinforcement decision-making model, the impact of different rehabilitation intervention paths on the long-term recovery effect is simulated, and the optimal value of each rehabilitation state is calculated, enabling the system to accurately evaluate the long-term effects of different intervention measures. By optimizing the rehabilitation strategy, calculating the optimal rehabilitation path, and combining the adaptive update mechanism of the personalized adjustment unit, the patient's recovery progress is evaluated in real time by using dynamic causal reasoning to ensure that the intervention strategy can be dynamically optimized as the patient's condition changes. The present invention improves the accuracy of the rehabilitation path decision-making, makes the rehabilitation plan for stroke patients more personalized, intelligent and dynamic, optimizes the rehabilitation effect, and at the same time reduces ineffective or inefficient medical interventions and improves the utilization rate of medical resources.

[0126] In this embodiment, for the information management method of stroke patients in the department of neurology, the method includes the following steps:

[0127] S71. Collect imaging data, physiological signals, genetic information and electronic medical record data of stroke patients in the Department of Neurology, convert and normalize the data format, remove redundant and abnormal data, and construct a structured multimodal data set;

[0128] S72. Based on structured multimodal data sets, a structural causal learning method is used to construct a causal model of disease evolution, identify variables and their causal relationships, generate a disease causal graph, eliminate confounding factors, and optimize the causal relationship matrix;

[0129] S73. Based on the disease causal graph, a self-supervised cross-modal representation learning method is used to extract shared features of structured multimodal datasets, perform cross-modal feature alignment and multimodal mapping, and generate a patient health profile.

[0130] S74. Based on the patient's health profile and disease causal graph, combined with the temporal causal reasoning method, predict the impact of different intervention measures on the development of the disease, and use reinforcement learning methods to optimize the disease intervention measures and form an intervention plan;

[0131] S75. Perform cross-modal consistency check on the disease causal diagram, patient health profile and intervention plan, use adaptive weighting strategy to adjust data contribution, and generate rehabilitation path decision data;

[0132] S76. Based on the rehabilitation pathway decision data, combined with the causal dynamic programming method, simulate the long-term impact of different intervention plans, calculate the optimal rehabilitation pathway, and dynamically adjust the intervention plan during the implementation of the rehabilitation pathway.

[0133] Embodiment 1:

[0134] In order to verify the feasibility of the present invention, the present invention is applied to the stroke patient management system of the Department of Neurology of a tertiary hospital. The hospital admitted a large number of patients with acute ischemic stroke and cerebral hemorrhage, with complicated conditions and long rehabilitation cycles. The traditional management method mainly relies on the experience of doctors and single data analysis, lacks personalized and precise intervention, resulting in large differences in patient rehabilitation effects. Some patients fail to adjust the rehabilitation plan in time, resulting in limited functional recovery or increased risk of recurrence. For this reason, the hospital decided to introduce the information management method for stroke patients in the Department of Neurology of the present invention to improve the precision and intelligence level of patient management and achieve personalized intervention and dynamic rehabilitation optimization.

[0135] In this system, the hospital establishes individualized health records for all stroke patients. During the hospitalization and rehabilitation processes, multi-modal data acquisition devices are used to obtain the patients' imaging data, physiological signals, genetic information, and electronic medical record data. All data is updated in real time through the hospital information management system and sensing devices and uploaded to the database of the stroke management system. After being standardized, the data enters the causal inference analysis module, which identifies the causal relationships between pathological feature variables based on causal learning methods, generates a causal diagram of the condition, and automatically eliminates confounding factors to improve the accuracy of condition analysis.

[0136] In the condition assessment phase, the system extracts the shared features of multi-modal data based on cross-modal learning methods to form an individualized health profile, and combines temporal causal inference methods to predict the trend of condition evolution. For different patients, the system can dynamically evaluate the post-stroke neurological function recovery path and optimize the individualized rehabilitation intervention plan in combination with the reinforcement learning algorithm. For example, the system compares the long-term effects of different intervention measures, recommends the optimal rehabilitation path, and adjusts according to the patients' real-time data during the rehabilitation process, so that the rehabilitation plan always meets the patients' dynamic recovery needs.

[0137] To verify the actual effect of the present invention, the hospital selected 400 stroke patients for a comparative experiment. Among them, 200 patients adopted the traditional stroke management method, and 200 patients adopted the intelligent management method of the present invention. The experimental period was 6 months. The performance of the two management methods in terms of rehabilitation progress, functional recovery, and optimization of medical interventions was compared. The experimental data is shown in the following table:

[0138] Table 1 Comparison of the effects of the traditional stroke management method and the method of the present invention

[0139]

[0140] It can be seen from the experimental data that the intelligent stroke management method of the present invention significantly improves the rehabilitation effect of patients. The increase in the Barthel index is 45.8% higher than that of the traditional method, indicating that the patients' ability to perform daily living activities recovers faster. Since the system can automatically optimize the rehabilitation path, the length of hospital stay is reduced by 4.6 days, improving the bed turnover rate of the hospital. The dynamic adjustment of the individualized rehabilitation plan increases the matching degree of the intervention plan from 74.6% to 91.8%, and the frequency of rehabilitation plan adjustment increases by 3.1 times, making the patients' rehabilitation strategies more targeted and adaptable. In addition, the system can respond to the changes in the patients' conditions in real time, shorten the intervention response time, reduce the recurrence rate of stroke, and improve the patients' rehabilitation satisfaction.

[0141] In the specific application process, the system provides precise interventions according to the individualized needs of different patients. For example, a 65-year-old male patient had hemiplegia after a stroke. The initial rehabilitation plan recommended 45 minutes of lower limb rehabilitation training per day. However, through time series analysis, the system found that the patient's muscle strength recovered quickly. Therefore, in the third week of rehabilitation, the plan was adjusted to increase the walking training time and combine it with nerve stimulation therapy. Another 58-year-old female patient had swallowing dysfunction after a stroke. Through multimodal analysis combined with causal reasoning, the system found that her recovery was slow and automatically adjusted the nutritional intervention strategy, recommending additional swallowing rehabilitation training, enabling her swallowing function to basically recover after 8 weeks.

[0142] In summary, through real-time data collection, causal reasoning analysis, cross-modal learning, construction of personalized health portraits, reinforcement learning optimization, and dynamic adjustment of the rehabilitation path, the present invention realizes the precise management of stroke patients, improves the rehabilitation efficiency and the utilization rate of medical resources. The system can dynamically adapt to the changes in the patient's condition, provide a personalized, intelligent, and optimized rehabilitation management plan for stroke patients, thereby improving the rehabilitation quality of patients and reducing the recurrence rate of stroke, providing an efficient and scientific stroke management model for medical institutions.

[0143] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its inventive concept, makes equivalent substitutions or changes, and should be covered within the protection scope of the present invention.

Claims

1. An information-based management system for stroke patients in the department of neurology, characterized in that, It includes the following modules: The data acquisition module collects the image data, physiological signals, genetic information, and electronic medical record data of stroke patients in the department of neurology, performs standardized processing and storage, removes redundant information, and constructs a structured multi-modal data set; The causal inference analysis module, based on the structured multi-modal data set, uses the structural causal learning method to analyze pathological features, eliminates confounding factors, constructs a causal model of disease evolution, and generates a causal graph of the disease; The cross-modal learning module, based on the causal graph of the disease, uses the self-supervised cross-modal representation learning method to extract the shared features of the structured multi-modal data set, performs feature alignment and multi-modal mapping, and generates a patient health portrait; The disease evolution analysis module, based on the patient health portrait and the causal graph of the disease, combines temporal causal inference to evaluate the impact of different intervention measures on the development of the disease and forms an intervention plan; The data fusion module performs cross-modal consistency verification on the causal graph of the disease, the patient health portrait, and the intervention plan, adopts an adaptive weighted strategy to adjust the data contribution degree, and generates rehabilitation path decision-making data; The rehabilitation optimization decision module, based on the rehabilitation path decision-making data, uses the causal dynamic programming method to simulate the long-term impact of different intervention plans and dynamically adjusts the intervention plan during the execution of the rehabilitation path; The causal inference analysis module specifically includes: The data parsing unit receives the structured multi-modal data set, including image data, physiological signals, genetic information, and electronic medical record data, performs modal separation on the structured multi-modal data set, extracts time series features and static features, and generates a standardized pathological data matrix; A causality modeling unit constructs a causal model of the disease progression based on a standardized pathological data matrix using a structural causal learning method, identifies the causal relationships between variables, and generates a causal association matrix , where represents the causal influence weight of variable on , and variables and variable respectively represent variables related to the disease The confounding factor elimination unit, based on the causal association matrix, uses the causal intervention inference method to calculate the direct causal effect between different variables to eliminate confounding factors and generate an adjusted causal association matrix: ； Among them, After excluding the influence of confounding factors under the constraint of the causal association matrix , the causal effect of variable on the disease state ; is the probability of the disease state under the combined action of variables, confounding factors and the causal association matrix; is the prior probability distribution of the confounding factor under the constraint of the causal association matrix ; The disease condition evolution modeling unit, based on the adjusted causal association matrix , combines time series analysis methods to calculate the changing trend of the disease condition and establish a time-series causal relationship network to evaluate the disease development path; A causal graph generation unit constructs a disease causal graph based on the disease evolution causal model. , where is a set of variables, is the causal association relationship, and the causal association relationship is the time-dependent causal relationship of the time series causal relationship network. The disease causal graph is transmitted to the cross-modal learning module. The disease evolution analysis module specifically includes: A disease state modeling unit, based on the patient's health profile Extract an individual pathological feature vector , construct a disease feature space, combine with the time-series causal reasoning method, calculate the causal association weights between different pathological features, and quantify the evolution process of the disease at continuous time steps; An intervention plan generation unit constructs a set of intervention measures based on the disease condition feature space , where is an intervention measure,[[]] is the type of the intervention measure, and an intervention measure impact function is defined to calculate the impact values of different intervention measures on the individual pathological feature vector and construct an intervention measure evaluation matrix :​ ； Among them, used to measure the intervention on the individual pathological feature vector contribution degree of, ; Reinforcement learning optimization unit, based on the intervention measure evaluation matrix , uses the reinforcement learning method to optimize the intervention measures, calculates the long-term return value of the intervention measures, and selects the optimal disease intervention path through the decision optimization function :[[]]END]] ； Among them, represents the long-term return value obtained by applying an intervention measure under the individual pathological feature vector, which is the time window for the evaluation of the intervention measure; An intervention path update unit, based on the individual pathological feature vector and the long-term return value of the intervention measure, dynamically adjusts the intervention measure to form an intervention plan.

2. The information management system for stroke patients in the department of neurology according to claim 1, wherein The confounding factor elimination unit specifically includes: Confounding factor identification unit, based on the causal association matrix , identify confounding factors , adopt the conditional independence test method to evaluate the variables and the disease status between the independence, by calculating the conditional independence probability to determine the confounding factor whether it has an impact on the causal relationship between the variable and the disease status ; A causal effect calculation unit, based on a causal association matrix and confounding factors , uses a causal intervention inference method to calculate the direct causal effect of variable on the disease state , eliminates the influence of confounding factors , and generates a causal effect calculation result after eliminating confounding factors; An adjustment matrix generation unit generates an adjusted causal association matrix based on the calculation result of the causal effect after removing confounding factors. , where represents removing confounding factors , and after that, the Bayesian update method is used to optimize the calculation of the adjusted causal association matrix. .

3. The information management system for stroke patients in the department of neurology according to claim 1, wherein, The cross-modal learning module specifically includes: Cross-modal feature extraction unit, receiving the disease causal graph , constructing a standardized feature matrix based on the structured multi-modal dataset ; Adopt a self-supervised cross-modal representation learning method to construct a cross-modal feature mapping model , and perform feature encoding on the standardized feature matrix to extract potential shared feature representations : ； Among them, is the potential shared feature representation, is the cross-modal feature mapping model, is the disease causal graph; A feature alignment unit, based on the latent shared feature representation , calculates the feature similarity between different modality data, and is optimized by a cross-modal contrastive learning loss function, so that the feature distribution distance of different modality features of the same patient in the latent shared feature representation is minimized, and the feature distribution distance of different patients is maximized, and generates an optimized shared feature representation ; A healthy image generation unit generates a patient's healthy image based on the optimized shared feature representation and adopts a weighted feature fusion strategy : ； Among them, is the feature fusion weight, is the fusion bias term, is the patient's health portrait.

4. The information management system for stroke patients in the department of neurology according to claim 1, wherein The rehabilitation optimization decision module specifically includes: A rehabilitation status modeling unit that receives rehabilitation path decision data and constructs a set of rehabilitation statuses , where represents the rehabilitation status of the patient at time moment; Causal dynamic programming unit, based on the set of rehabilitation states and the set of intervention measures , using the causal dynamic programming method to simulate the impact of different rehabilitation paths on the long-term recovery effect, define the optimal value function of the rehabilitation path , and calculate the optimal value of the rehabilitation path for each rehabilitation state : ； in, In recovery state Implement intervention measures The immediate recovery benefits obtained, is the discount factor, which measures the balance between current rehabilitation benefits and long-term rehabilitation impacts. is the patient's recovery status at the next moment after the intervention, Indicates recovery status The optimal value of the rehabilitation path; The rehabilitation path optimization unit calculates the optimal rehabilitation path based on the optimal values of the rehabilitation paths in each rehabilitation state; The personalized adjustment unit uses an adaptive update mechanism to adjust the optimal rehabilitation path, evaluates the patient's recovery progress through dynamic causal inference, optimizes the intervention plan, and makes the intervention plan adapt to the individual recovery needs.

5. The information-based management method for stroke patients in the department of neurology, which is applied to the information-based management system for stroke patients in the department of neurology described in claims 1-4, is characterized in that It includes the following steps: Collect the image data, physiological signals, genetic information, and electronic medical record data of stroke patients in the department of neurology, perform data format conversion and normalization processing, remove redundant and abnormal data, and construct a structured multi-modal data set; Based on the structured multi-modal data set, use the structural causal learning method to construct a causal model of disease evolution, identify variables and their causal association relationships, generate a causal graph of the disease, and eliminate confounding factors to optimize the causal association matrix; Based on the causal graph of the disease, use the self-supervised cross-modal representation learning method to extract the shared features of the structured multi-modal data set, perform cross-modal feature alignment and multi-modal mapping, and generate a patient health portrait; Based on the patient health portrait and the causal graph of the disease, combine the temporal causal inference method to predict the impact of different intervention measures on the development of the disease, and use the reinforcement learning method to optimize the disease intervention measures to form an intervention plan; Perform cross-modal consistency verification on the disease causality diagram, patient health portrait, and intervention plan, adopt an adaptive weighted strategy to adjust the data contribution degree, and generate rehabilitation path decision-making data; Based on the rehabilitation path decision-making data, combine the causal dynamic programming method to simulate the long-term effects of different intervention plans, calculate the optimal rehabilitation path, and dynamically adjust the intervention plan during the execution of the rehabilitation path.

Images