Cerebral stroke dysfunction refined model construction method and system based on ICF theory

Abstract

The invention belongs to the technical field of medical care informatics, particularly relates to a method and a system for constructing a refined model of cerebral apoplexy dysfunction based on an ICF theory, and aims to construct a refined model capable of accurately describing a dysfunction area of a cerebral apoplexy patient through dynamic optimization and a self-adaptive adjustment mechanism. In order to improve the accuracy and efficiency of dysfunction assessment, multi-modal assessment data, including pathological information, imaging data and biochemical indexes, of a patient is continuously collected and analyzed through a dynamic scoring mechanism, then, in combination with an ICF theory, feature parameters of a barrier region are automatically extracted and optimized through feature matching and a deep learning algorithm, and the accuracy and efficiency of dysfunction assessment are improved. According to the method, a three-dimensional model capable of accurately evaluating and positioning an obstacle area is constructed, and the model not only can update and adaptively adjust parameters in real time, but also can provide personalized rehabilitation suggestions and treatment schemes according to individual differences of different patients.

Description

Technical Field

[0001] The present invention belongs to the technical field of healthcare informatics, and particularly relates to a method and system for constructing a refined model of stroke dysfunction based on the ICF theory. Background Art

[0002] Stroke is a complex disease with multiple pathophysiological injuries caused by the interaction of various pathophysiological injury factors. It is also known as a stroke. Severe cases can lead to death, and as the course of stroke prolongs, the number of stroke deaths also increases accordingly. Therefore, timely detecting the dysfunction areas of stroke patients and performing corresponding physical therapy on these dysfunction areas can greatly reduce the death probability of patients.

[0003] In the prior art, the ICF theory (International Classification of Functioning, Disability and Health) is the most suitable function classification system for the modern rehabilitation medicine model. By adopting an internationally unified standardized language, it provides terms, definitions, and classifications related to functional health and disability. Currently, the ICF theory has been widely applied in many fields such as health monitoring, function assessment, treatment planning, health care, disability welfare distribution, vocational assessment, social policy formulation, universal design, education programs, income insurance, and intervention management. However, in the process of application, the existing ICF theory cannot accurately describe stroke dysfunction, which is likely to result in inaccurate descriptions of the area and location of dysfunction, thereby leading to improper treatment of the patient's dysfunction area, increasing the patient's death probability or prolonging the patient's recovery period. Therefore, a refined model of the stroke dysfunction area based on the ICF theory has emerged. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for constructing a refined model of stroke dysfunction based on the ICF theory, aiming to construct a refined model that can accurately describe the dysfunction area of stroke patients through a dynamic optimization and adaptive adjustment mechanism, so as to improve the accuracy and efficiency of dysfunction assessment.

[0005] The technical solutions adopted by the present invention are specifically as follows: A method for constructing a refined model of stroke dysfunction based on the ICF theory, comprising: Obtaining multi-modal assessment data of stroke patients and performing preprocessing, where the multi-modal assessment data includes patient medical record information, function assessment reports, and medical staff feedback information; Performing cluster analysis on the preprocessed multi-modal assessment data to obtain classification information, where the classification information includes dysfunction injury information and non-dysfunction injury information; Performing feature extraction based on the dysfunction injury information, outputting the extracted features, and summarizing them into a feature label set; Encode each feature label in the feature label set to obtain reference encoding data; Construct a mapping model based on the correspondence between the reference encoding data and the extracted features, map new input data through the mapping model, and dynamically optimize it according to the application effect of the mapping model to continuously improve its mapping accuracy. Specifically, the steps corresponding to the dynamic optimization include: Obtain the current application effect of the mapping model and determine the current optimization amplitude based on this; Real-time monitor the user feedback during the application process of the mapping model, and analyze and process it to obtain optimization points; Construct an optimization plan according to the optimization points and the current optimization amplitude, and optimize the mapping model according to the optimization plan; Classify the dynamic optimization according to the application effect of the optimization plan, bind the level of the dynamic optimization to the optimization parameters, and synchronously store each optimization parameter.

[0006] In a preferred scheme, the steps of obtaining the multi-modal evaluation data of stroke patients and performing preprocessing include: Collect the multi-modal evaluation data of stroke patients and perform preprocessing on it; Perform standardization processing on the multi-modal evaluation data to unify its format; Extract the feature information of the multi-modal evaluation data, and perform duplicate removal and noise reduction processing on the feature information; Perform pairing processing on the processed feature information to obtain pairing parameters, and classify each pairing parameter into a reference input condition and a post-input condition; Synchronously process the reference input condition and the post-input condition and output them as a reference condition parameter set.

[0007] In a preferred scheme, when performing duplicate removal and noise reduction processing on the feature information of the multi-modal evaluation data, construct a noise filtering model, input the extracted feature information into the noise filtering model to identify invalid or incorrect data in the feature information, and add an anomaly monitoring module to the noise filtering model to monitor the extraction process of the feature information in real time, and judge whether the extraction process of the feature information is abnormal in combination with the monitoring data output by the anomaly monitoring module.

[0008] In a preferred scheme, before the step of synchronously processing the reference input condition and the post-input condition and outputting them as a reference condition parameter set, calculate the course length according to the feature information, and determine the combination method of the reference input condition and the post-input condition according to the course length.

[0009] In a preferred embodiment, after the reference condition parameter set is output, condition parameter matching is performed based on the reference input condition and the post-input condition in the reference condition parameter set, and then parameters matching the condition parameters are screened according to the disease course length and output as a preprocessing result.

[0010] In a preferred embodiment, after the preprocessing result is output, it is used as an input to access a condition parameter correction model for correction. In the condition parameter correction model, a condition parameter correction database is constructed. The condition parameter correction database is provided with a condition pre-subset, a condition post-subset, and a condition parallel subset. Among them, multiple condition parameters are preset in the condition pre-subset, the condition post-subset, and the condition parallel subset.

[0011] In a preferred embodiment, the step of extracting features based on the dysfunction injury information, outputting the extracted features, and summarizing them into a feature label set includes: Obtaining multi-modal evaluation data corresponding to the dysfunction injury information; Performing feature extraction on the multi-modal evaluation data to obtain multiple injury features; Scoring the injury features, merging the injury features with close scores, summarizing them into extracted features, then marking the extracted features with extraction labels, and outputting a feature label set; By constructing a dynamic scoring mechanism, continuously collecting injury features, and continuously updating the feature scores of the scoring mechanism to improve the accuracy of feature extraction.

[0012] In a preferred embodiment, after the step of dynamic optimization, the application effect of the optimization plan is evaluated and analyzed to obtain an application score; Sorting the optimization plans based on the application score, and selecting the historical optimization plans of the TOP-K optimization plans for combined application to generate an optimal optimization plan; Optimizing the mapping model according to the optimal optimization plan; Synchronously monitoring the optimization effect of the optimal optimization plan on the mapping model.

[0013] In a preferred embodiment, after optimizing the mapping model according to the optimal optimization plan, by introducing a supervised learning mechanism, adding manual intervention measures, adjusting the output result of the mapping model to make it more in line with the actual needs, and real-time collecting user feedback, refining optimization suggestions from it, and adjusting the mapping model parameters according to the optimization suggestions.

[0014] The present invention also provides a refined model construction system for stroke dysfunction based on the ICF theory, which is applied to the above-mentioned method for constructing a refined model of stroke dysfunction based on the ICF theory, including: A data acquisition module, which is used to obtain multi-modal evaluation data of stroke patients and perform preprocessing. The multi-modal evaluation data includes patient medical record information, functional evaluation reports, and medical staff feedback information; A data analysis module, which is used to perform cluster analysis on the preprocessed multi-modal evaluation data to obtain classification information. Among them, the classification information includes functional disorder injury information and non-functional disorder injury information; A feature extraction module, which is used to extract features based on the functional disorder injury information, output the extracted features, and summarize them into a feature label set; An encoding module, which is used to perform encoding processing on each feature label in the feature label set to obtain reference encoding data; A model construction module, which is used to construct a mapping model based on the corresponding relationship between the reference encoding data and the extracted features, and map new input data through the mapping model, and dynamically optimize it according to the application effect of the mapping model, continuously improving its mapping accuracy.

[0015] Beneficial effects: The present invention constructs a refined model that can accurately describe the functional disorder area of stroke patients through a dynamic optimization and adaptive adjustment mechanism. Specifically, by introducing a dynamic scoring mechanism, continuously collecting and analyzing the multi-modal evaluation data of patients, including pathological information, imaging data, and biochemical indicators. Secondly, combined with the ICF theory, through feature matching and deep learning algorithms, automatically extract and optimize the feature parameters of the disorder area, and construct a three-dimensional model that can accurately evaluate and locate the disorder area. This model can not only be updated in real time and adaptively adjust parameters, but also provide personalized rehabilitation suggestions and treatment plans according to the individual differences of different patients. Finally, by introducing a feedback closed-loop mechanism, continuously collecting and analyzing the application effects, continuously optimizing and adjusting the model to improve the accuracy and efficiency of evaluation, making the accuracy of functional disorder evaluation higher and the efficiency faster. Description of the Drawings

[0016] Figure 1 It is a schematic flow chart of the method of the present invention; Figure 2 It is a schematic diagram of the sub-steps of step S1 of the present invention; Figure 3 It is a schematic diagram of the sub-steps of step S3 of the present invention; Figure 4 It is a schematic diagram of the sub-steps of step S6 of the present invention; Figure 5 It is a schematic structural diagram of the system of the present invention. Detailed Embodiments

[0017] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will describe the technical solutions in the present invention or the prior art in conjunction with the accompanying drawings and specific embodiments.

[0018] Embodiment 1 Please refer to the attached Figure 1 , attached Figure 1 is a flowchart of a method provided based on an embodiment of the present invention, which includes: S1. Obtain multimodal assessment data of stroke patients and perform preprocessing. The multimodal assessment data includes patient medical record information, functional assessment reports, and medical staff feedback information; S2. Perform clustering analysis on the preprocessed multimodal assessment data to obtain classification information. Among them, the classification information includes functional impairment information and non-functional impairment information; S3. Extract features based on the functional impairment information, output the extracted features, and summarize them into a feature label set; S4. Perform encoding processing on each feature label in the feature label set to obtain reference encoding data; S5. Construct a mapping model based on the corresponding relationship between the reference encoding data and the extracted features, and map new input data through the mapping model. And dynamically optimize it according to the application effect of the mapping model, and continuously improve its mapping accuracy.

[0019] Specifically, the steps corresponding to the dynamic optimization include: Obtain the current application effect of the mapping model and determine the current optimization amplitude based on this; Real-time monitor the user feedback during the application process of the mapping model, and analyze and process it to obtain optimization points; Construct an optimization plan according to the optimization points and the current optimization amplitude, and optimize the mapping model according to the optimization plan; Perform hierarchical processing on the dynamic optimization according to the application effect of the optimization plan, bind the level of the dynamic optimization with the optimization parameters, and synchronously store each optimization parameter.

[0020] As described in steps S1 - S5, in the method for constructing a refined model of stroke dysfunction based on the ICF theory in this embodiment, first, multi-modal evaluation data of stroke patients is obtained, including patient medical record information, functional evaluation reports, and feedback information from medical staff, etc. Then, these data are standardized and duplicate data is removed to ensure the accuracy and consistency of the data. Next, the preprocessed data is subjected to cluster analysis to classify the dysfunction injury information and non-dysfunction injury information. This classification result can help better understand the patient's dysfunction condition. Then, feature extraction is performed based on the dysfunction injury information, and a feature label set is output. To better process these feature labels, they are encoded to obtain benchmark encoded data. These encoded data can form the basis for subsequent model construction. In the model construction stage, a mapping model needs to be constructed according to the corresponding relationship between the benchmark encoded data and the extracted features. Through this mapping model, new input data can be mapped to form a refined model and clearly show the patient's dysfunction, so as to help doctors make corresponding diagnoses more quickly and avoid regional confirmation errors caused by individual ability differences.

[0021] Finally, in this embodiment, by introducing a dynamic optimization mechanism, the mapping accuracy of its optimized mapping model is continuously improved. Specifically, by real-time monitoring of user feedback, analyzing user behavior and their evaluation of the mapping model, optimization points can be collected. With this support, this embodiment will construct corresponding optimization plans to guide the improvement process of the mapping model. An important point in this embodiment is that, according to the application effect of the optimization plan, the dynamic optimization is classified, and the corresponding dynamic optimization parameters are bound to the mapping model to coordinate the relationship between various dynamic optimizations.

[0022] Specifically, as Figure 2 shown, the steps of obtaining multi-modal evaluation data of stroke patients and performing preprocessing include: S11. Collect multi-modal evaluation data of stroke patients and perform preprocessing on it; S12. Standardize the multi-modal evaluation data to unify its format; S13. Extract the feature information of the multi-modal evaluation data and perform duplicate data removal and noise removal processing on the feature information; S14. Perform pairing processing on the processed feature information to obtain pairing parameters, and classify each of the pairing parameters into a benchmark input condition and a post-input condition; S15. Synchronize the benchmark input condition and the post-input condition and output them as a set of benchmark condition parameters.

[0023] To comprehensively understand the health status of stroke patients, it is first necessary to collect and preprocess relevant data. Starting from this step, a large amount of multimodal assessment data is collected, including the basic information, medical history, physical examination reports, and various medical imaging data of stroke patients, to ensure the comprehensiveness and accuracy of the data.

[0024] Then, careful and systematic preprocessing work is carried out on the multimodal assessment data to exclude any potential data noise, redundancy, or anomalies, to ensure the purity and consistency of the data. Next, the multimodal assessment data is standardized to unify its format for better subsequent analysis. During this process, all data is converted into a standard format. For example, it is ensured that all numerical values use a unified unit, time format, and naming convention, so that calculation errors will not occur due to format when processing in different environments.

[0025] After that, feature extraction is performed to extract important feature information. For example, different types of data and information of different natures are extracted to further simplify the data used. In addition, the feature information of the multimodal assessment data needs to be de-duplicated and de-noised to ensure the uniqueness and purity of the data. Then, pairing processing is carried out to obtain pairing parameters, and each pairing parameter is classified into a reference input condition and a post-input condition to help clarify the relationship between the data. For example, the dependency relationship and constraint conditions between different parameters are determined. Finally, the reference input condition and the post-input condition are synchronously processed and summarized and output as a set of reference condition parameters. With the help of modern information technology, the above steps are automatically completed to form a complete and efficient data processing system.

[0026] Specifically, for the step of classifying each pairing parameter into a reference input condition and a post-input condition, when de-duplicating and de-noising the feature information of the multimodal assessment data, a noise filtering model is constructed, and the extracted feature information is input into the noise filtering model to identify invalid or incorrect data in the feature information. An anomaly monitoring module is added to the noise filtering model to monitor the extraction process of the feature information in real time, and it is judged whether the extraction process of the feature information is abnormal by combining the monitoring data output by the anomaly monitoring module. During the implementation of the noise filtering model, a corresponding data filtering module is set up to help identify invalid or incorrect data, thus avoiding the appearance of chaotic data and preventing misleading effects on subsequent analysis. At the same time, an anomaly monitoring module is added to help monitor the extraction process of the feature information in real time to ensure the accuracy of the extracted feature information. And because the sources of the multimodal assessment data used in the feature extraction process are diverse, this step introduces an anomaly monitoring module to also identify in real time whether there are problems with data extraction errors.

[0027] Furthermore: After synchronizing the reference input conditions and the post-input conditions and outputting them as a reference condition parameter set, calculate the disease course length according to the characteristic information, and determine the combination method of the reference input conditions and the post-input conditions according to the disease course length.

[0028] During the synchronization process of the reference input conditions and the post-input conditions, calculating the disease course length based on the characteristic information can determine the combination method of the reference input conditions and the post-input conditions. By considering the performance differences of the lesion areas under different disease courses, a connection is established between the basic input conditions and the post-term conditions, ensuring the output order between the reference input conditions and the post-input conditions.

[0029] Specifically, after the output of the reference condition parameter set, perform condition parameter matching based on the reference input conditions and the post-input conditions in the reference condition parameter set, then screen the parameters that match the condition parameters according to the disease course length, and output them as the preprocessing result.

[0030] Among them, by importing the corresponding reference input conditions and post-input conditions, based on the preset disease course length, screen and match the reference input conditions and the post-input conditions, output the key areas that have a greater impact on the lesion area, and clearly display the cerebrovascular circulation system of stroke patients, locate the lesion area, so as to help doctors respond more quickly and make correct treatments.

[0031] Specifically, after the output of the preprocessing result, use it as the input and connect it to the condition parameter correction model for correction. In the condition parameter correction model, a condition parameter correction database is constructed. In the condition parameter correction database, a condition pre-subset, a condition post-subset, and a condition parallel subset are set. Among them, multiple condition parameters are preset in the condition pre-subset, the condition post-subset, and the condition parallel subset. In this step, the preprocessing result will be used as the input, and the preprocessing result will be further adjusted and optimized by connecting to a condition parameter correction model with rich models. In the personal condition parameter correction model, first create a condition parameter correction database, and then create a condition pre-subset, a condition post-subset, and a condition parallel subset according to the execution order of the condition parameters, and then call the condition parameters according to the execution order and output the corresponding post-condition parameters to provide support for further optimizing the preprocessing result.

[0032] Specifically, as Figure 3 shown, the steps of feature extraction based on the functional impairment information, outputting the extracted features, and summarizing them into a feature label set include: S31. Obtain the multi-modal evaluation data corresponding to the functional impairment information; S32. Extract features from the multi-modal evaluation data to obtain multiple damage features; S33. Score the damage features, merge the damage features with close scores, summarize them into extracted features, then label the extracted features with extraction labels and output them as a feature label set; By constructing a dynamic scoring mechanism, continuously collect damage features and continuously update the feature scores of the scoring mechanism to improve the accuracy of feature extraction.

[0033] As described in steps S31 - S33, when extracting features based on dysfunction damage information, first, the multi-modal evaluation data corresponding to the corresponding dysfunction damage information is acquired by the system, and feature extraction is performed through deep learning and image processing techniques. After output, it is matched with the damage features to help better understand the nature and location of the damage features, and then further evaluate and analyze the dysfunction, thereby evaluating the degree of dysfunction damage of stroke to the patient, giving doctors more definite treatment opinions, and ensuring that patients receive timely and appropriate physical therapy.

[0034] Based on the matching relationship between the multi-modal evaluation data and the damage features, the matched multi-modal evaluation data is scored to output a damage feature score table to rank each damage feature, corresponding to the priority of the damage feature in order, and the damage features with low priority are merged into extracted features. Then, the extracted features are labeled as extraction labels and saved in the feature label set to further correspond to the priority of the extraction label according to the priority of the damage feature, thus forming a clear, detailed, and highly logical feature label set.

[0035] By constructing a dynamic scoring mechanism, continuously collect damage features and convert them into feature scores to update the damage feature score table, so as to improve the accuracy and efficiency of feature extraction, enable doctors to quickly and accurately obtain the optimal extracted features, respond to the patient's condition in a timely manner, and perform corresponding physical therapy on the patient in a timely manner when necessary, shortening the response cycle.

[0036] In the method for constructing a refined model of stroke dysfunction based on the ICF theory, the average extraction accuracy of the median features of the index multi-modal evaluation data and the damage features reaches 34.07%. Among them, for the extracted features with equal scores, the priorities are the same. However, after the scoring requirements of each extracted feature are met, the sorting priority can further improve the priority of the current extracted feature. Based on this, a dynamic scoring mechanism is further introduced for continuous optimization, doubling the average extraction accuracy by nearly 6%, and it can also merge multiple extracted features with relatively low extraction accuracy into one extracted feature, increasing the comprehensiveness of stroke evaluation. In this way, the extraction label set is continuously collected and updated to provide a more accurate auxiliary solution for the follow-up, improving the doctor's response speed and decision-making correctness.

[0037] Specifically, as Figure 4 shown: After dynamically optimizing the application effect of the mapping model, evaluate and analyze the application effect of the optimization plan to obtain an application score; S61. Sort the optimization plans based on the application score, and select the historical optimization solutions of the TOP-K optimization plans for combined application to generate an optimal optimization plan; S62. Optimize the mapping model according to the optimal optimization plan; S63. Synchronously monitor the optimization effect of the optimal optimization plan on the mapping model.

[0038] As described in steps S61 - S63, after dynamically optimizing the application effect of the mapping model, first, evaluate and analyze the application effect of each optimization plan and summarize it in a unified evaluation and analysis form. Then, sort the application scores in the optimization plan list, select the TOP-K optimization plans with the best optimization effect, and perform combined application on their historical optimization solutions to generate a highly feasible optimal plan. Then, optimize the mapping model according to the optimal plan, and synchronously monitor the optimization effect of the optimal optimization plan on the mapping model by combining various feedback means, and make further adjustments and optimizations as needed.

[0039] Specifically, in the historical optimization solutions, comprehensively consider the success rate and stability of each optimization solution in historical applications, select those standardized low-score optimization solutions that have been fully verified, and then, according to the priority information of the lesion damage characteristics of the doctor, give different standardized low-score value ranges for the damage characteristics of different lesion types to generate an optimal optimization plan, and extract it as needed into the subsequent mapping plan to improve the optimization efficiency of the mapping model.

[0040] Specifically, after optimizing the mapping model according to the optimal optimization plan, by introducing a supervised learning mechanism, increasing manual intervention measures, adjusting the output results of the mapping model to make it more in line with real-world requirements, and collecting user feedback in real time, refining optimization suggestions from it, and adjusting the mapping model parameters according to the optimization suggestions.

[0041] The mapping model continuously receives various types of constraint condition information and outputs different types of mapping results, prohibiting the use of other types of mapping models without explicit authorization. However, if the expansion of the mapping model category is too fast, it will lead to a phenomenon of abnormal matching of mapping result parameters between different types of mapping models, thus echoing mapping result types that do not match the original drawing. Obviously, this deviates from the original intention of the user and is relatively serious.

[0042] In this embodiment, by combining the positive and negative feedback of users under each mapping model (such as basic feedback information like smiling faces and crying faces, and the application of detailed evaluation information should also be combined later), the expansion of the mapping model categories is curbed to avoid incorrect mapping result types. In the long run, user feedback is combined into the dynamic scoring link, enabling it to participate in the feedback and optimization of the accuracy of damage features synchronously. Every 25 pieces of user feedback, the present invention outputs the TOP-K mapping results of the mapping model, and further screens them through machine algorithms and human intervention, so as to efficiently present them to users in the form of graphic texts. And there has been a large number of industry practices proving that the display method of graphic texts not only improves the reading efficiency of users, but also increases their satisfaction. Next, the output mapping results are further docked to the next stage to meet the needs of users. Otherwise, the categories of mapping results are reduced to continuously improve the user experience. While improving user satisfaction, the present invention can also provide a set of practical industry display tools for the majority of scientific researchers, enabling scientific research results not to be limited to the boring text descriptions in papers, but to be presented in a more acceptable dynamic effect.

[0043] Embodiment 2 As Figure 5 shown, this embodiment provides a refined model construction system for stroke dysfunction based on the ICF theory, which is applied to the above-mentioned method for constructing a refined model of stroke dysfunction based on the ICF theory, and includes: A data acquisition module, which is used to obtain multi-modal evaluation data of stroke patients and perform preprocessing. The multi-modal evaluation data includes patient medical record information, functional evaluation reports, and medical staff feedback information; A data analysis module, which is used to perform cluster analysis on the preprocessed multi-modal evaluation data to obtain classification information. Among them, the classification information includes functional disorder damage information and non-functional disorder damage information; A feature extraction module, which is used to extract features based on the functional disorder damage information, output the extracted features, and summarize them into a feature label set; An encoding module, which is used to perform encoding processing on each feature label in the feature label set to obtain reference encoding data; A model construction module, which is used to construct a mapping model based on the corresponding relationship between the reference encoding data and the extracted features, and map new input data through the mapping model, and dynamically optimize it according to the application effect of the mapping model to continuously improve its mapping accuracy.

[0044] In the above system, the data acquisition module is used to obtain multi-modal assessment data of stroke patients from high-quality medical data sources to ensure the comprehensiveness and accuracy of the data. By using advanced data acquisition tools and technologies, it can obtain multi-modal assessment data from different medical facilities, devices, and environments, covering various information such as the patient's pathology, genes, images, biometric features, etc. After data acquisition, preprocessing of the multi-modal assessment data is performed, including data cleaning, deduplication, missing value filling, outlier handling, data transformation, normalization, etc., to ensure the quality and consistency of the data. And according to specific requirements and purposes, the data is mapped to a suitable representation space or stored and transmitted in a suitable form to increase the efficiency of the data acquisition process and ensure the quality, privacy, and security of the data.

[0045] The data analysis module is used to identify data patterns and characteristics of functional impairment in different stroke patients, find similar groups of functional impairments, sort out, summarize, and compress the multi-dimensional and complex information in the multi-modal assessment data, and is used to divide the multi-modal assessment data of stroke patients into different categories to generate classification information, including functional impairment information and non-functional impairment information, to help clarify the damaged and undamaged features of stroke patients in terms of function. Combining standardized data construction and label description matching to construct a visual and refined assessment graph to provide intuitive assistance to doctors.

[0046] The feature extraction module is used to automatically extract key features from the functional impairment information. For example, it extracts significant features such as the patient's physical signs, symptoms, and image data and outputs the extracted features, so as to improve the efficiency and accuracy of assessment and diagnosis, enabling doctors to quickly understand the patient's functional impairment situation, reducing the time loss caused by describing features, and increasing the patient's chance of receiving corresponding physical therapy.

[0047] The encoding module is used to encode each feature label in the feature label set output by the feature extraction module, convert it into a digital signal that is convenient for processing, storage, and transmission, and avoid information loss when encoding the feature labels to ensure that the encoded feature labels can accurately express the original multi-modal data. And store these encoded feature labels in the database so that the encoded feature labels can be transmitted and shared between different systems and platforms, providing support for subsequent data warehouses, data analysis, and data mining, etc.

[0048] The model construction module is used to construct a mapping model based on the corresponding relationship between the reference coding data in the coding module and the features extracted by the feature extraction module. After the mapping model is built, further mapping operations can be carried out to help convert the medical data or images of patients into a format that can be expressed in the mother tongue, so as to be expressed more accurately, enabling nurses, patients, and their families to have a preliminary understanding of stroke patients without the need for doctors to give too many explanations, allowing doctors to have more energy to confirm functional impairments.

[0049] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a refined stroke dysfunction model based on ICF theory, characterized by: include: Obtain multimodal assessment data of stroke patients and perform preprocessing; The multimodal assessment data includes patient medical record information, functional assessment reports, and medical staff feedback information; Performing cluster analysis on the pre-processed multimodal assessment data to obtain classification information, wherein the classification information includes functional disorder injury information and non-functional disorder injury information; Extract features based on the functional impairment information, output the extracted features, and summarize them into a feature label set; Encoding each feature tag in the feature tag set to obtain reference coded data; A mapping model is constructed according to the correspondence between the reference coded data and the extracted features, and new input data is mapped through the mapping model. The mapping model is dynamically optimized according to its application effect to continuously improve its mapping accuracy; Specifically, the steps corresponding to the dynamic optimization include: Obtaining the current application effect of the mapping model, and determining the current optimization range based on the current application effect of the mapping model; Real-time monitoring of user feedback during the application of the mapping model, and analysis and processing thereof to obtain optimization points; Constructing an optimization plan according to the optimization point and the current optimization range, and optimizing the mapping model according to the optimization plan; The dynamic optimization is graded according to the application effect of the optimization plan, and the grade of the dynamic optimization is bound to the optimization parameter, and each optimization parameter is stored synchronously.

2. The method for constructing a refined stroke dysfunction model based on ICF theory according to claim 1, characterized in that: The step of obtaining multimodal assessment data of stroke patients and performing preprocessing includes: Collect multimodal assessment data of stroke patients and preprocess them; Standardizing the multimodal evaluation data to unify its format; Extracting feature information of the multimodal evaluation data, and performing deduplication and denoising processing on the feature information; Performing pairing processing on the processed feature information to obtain pairing parameters, and classifying each pairing parameter into a reference input condition and a post-input condition; The reference input condition and the post-input condition are synchronously processed and output as a reference condition parameter set.

3. The method for constructing a refined stroke dysfunction model based on ICF theory according to claim 2, characterized in that: When the feature information of the multimodal evaluation data is deduplicated and denoised, a noise filtering model is constructed and the extracted feature information is input into the noise filtering model to identify invalid or erroneous data in the feature information. An abnormality monitoring module is added to the noise filtering model to monitor the extraction process of the feature information in real time, and whether the feature information extraction process is abnormal is determined in combination with the monitoring data output by the abnormality monitoring module.

4. The method for constructing a refined stroke dysfunction model based on ICF theory according to claim 3, characterized in that: Before the step of synchronously processing the benchmark input conditions and the post-input conditions and outputting them as a benchmark condition parameter set, the course length is calculated according to the characteristic information, and the combination method of the benchmark input conditions and the post-input conditions is determined according to the course length.

5. The method for constructing a refined stroke dysfunction model based on ICF theory according to claim 4, characterized in that: After the benchmark condition parameter set is output, condition parameter matching is performed based on the benchmark input conditions and post-input conditions in the benchmark condition parameter set, and then parameters matching the condition parameters are screened according to the course length and output as preprocessing results.

6. The method for constructing a refined stroke dysfunction model based on ICF theory according to claim 5, characterized in that: After the preprocessing result is output, it is used as input and connected to a conditional parameter correction model for correction. In the conditional parameter correction model, a conditional parameter correction database is constructed. The conditional parameter correction database is provided with a conditional pre-subset, a conditional post-subset and a conditional parallel subset, wherein a plurality of conditional parameters are preset in the conditional pre-subset, the conditional post-subset and the conditional parallel subset.

7. The method for constructing a refined stroke dysfunction model based on ICF theory according to claim 1, characterized in that: The step of extracting features based on the functional impairment information, outputting the extracted features, and summarizing them into a feature label set includes: Acquiring multimodal assessment data corresponding to the functional impairment information; Extracting features from the multimodal evaluation data to obtain multiple damage features; Scoring the damage features, merging the damage features with similar scores to extract features, marking the extracted features with extraction labels, and outputting a feature label set; By building a dynamic scoring mechanism, damage features are continuously collected, and the feature scores of the scoring mechanism are continuously updated to improve the accuracy of feature extraction.

8. The method for constructing a refined stroke dysfunction model based on ICF theory according to claim 1, characterized in that: After the step of dynamic optimization, the application effect of the optimization plan is evaluated and analyzed to obtain an application score; The optimization plans are sorted based on the application scores, and historical optimization solutions of the TOP-K optimization plans are selected for combined application to generate an optimal optimization plan; Optimizing the mapping model according to the optimal optimization plan; The optimization effect of the optimal optimization plan on the mapping model is simultaneously monitored.

9. The method for constructing a refined stroke dysfunction model based on ICF theory according to claim 8, characterized in that: After the mapping model is optimized according to the optimal optimization plan, a supervised learning mechanism is introduced and manual intervention measures are added to adjust the output result of the mapping model to make it more in line with actual needs. User feedback is collected in real time to extract optimization suggestions and adjust the mapping model parameters according to the optimization suggestions.

10. A system for constructing a refined stroke dysfunction model based on ICF theory, applied to the method for constructing a refined stroke dysfunction model based on ICF theory as claimed in any one of claims 1 to 9, characterized in that: include: A data acquisition module, which is used to obtain multimodal assessment data of stroke patients and perform preprocessing, wherein the multimodal assessment data includes patient medical history information, functional assessment reports, and feedback information from medical staff; A data analysis module, the data analysis module is used to perform cluster analysis on the pre-processed multimodal evaluation data to obtain classification information, wherein the classification information includes functional disorder injury information and non-functional disorder injury information; A feature extraction module, the feature extraction module is used to extract features based on the functional impairment information, output the extracted features, and summarize them into a feature label set; An encoding module, the encoding module is used to encode each feature tag in the feature tag set to obtain reference encoding data; A model building module, the model building module is used to build a mapping model based on the correspondence between the benchmark coding data and the extracted features, and to map new input data through the mapping model, and to dynamically optimize the mapping model according to its application effect, so as to continuously improve its mapping accuracy.

Images