Electronic medical record analysis and personalized rehabilitation recommendation method and system based on large model

Abstract

The invention discloses an electronic medical record analysis and personalized rehabilitation recommendation method and system based on a large model, and relates to the technical field of medical information processing and artificial intelligence. Comprising the steps of collecting multi-source medical data of a patient from a hospital information system; preprocessing the multi-source medical data; fusing the preprocessed multi-source medical data, and constructing a multi-modal patient portrait; inputting the constructed patient portrait into a pre-trained artificial intelligence analysis model, and performing multi-dimensional analysis to obtain an analysis result; constructing a personalized rehabilitation recommendation engine; based on the analysis result, generating a rehabilitation recommendation scheme through a personalized rehabilitation recommendation engine; in the execution process of the rehabilitation recommendation scheme, continuously collecting feedback data of the patient, and performing incremental learning and retraining on the artificial intelligence analysis model. According to the method, the electronic medical record can be automatically and accurately analyzed, and the personalized rehabilitation recommendation scheme is generated based on the analysis result.

Description

Technical Field

[0001] This invention relates to the fields of medical information processing and artificial intelligence technology, and in particular to a method and system for electronic medical record analysis and personalized rehabilitation recommendations based on a large model. Background Technology

[0002] Electronic medical record systems are now widely used in healthcare institutions, storing a vast amount of patients' clinical data, including diagnostic records, laboratory reports, imaging data, and medication records. This data is a valuable resource for patient health assessments and rehabilitation planning.

[0003] However, current utilization of electronic medical record data often results in unstructured natural language text or scattered structured data, making it difficult for doctors to quickly and comprehensively integrate and analyze, thus burying a large amount of valuable information. Traditional analysis methods mostly rely on keyword retrieval or simple statistics, lacking the ability to deeply explore and predict the severity of a patient's condition, the risk of complications, and prognosis. Current rehabilitation recommendations are mostly based on clinical pathways or doctors' personal experience, lacking refined and dynamically adjusted personalized rehabilitation plans tailored to individual patient characteristics, leading to poor rehabilitation outcomes or wasted resources.

[0004] Therefore, proposing a method and system for electronic medical record analysis and personalized rehabilitation recommendations based on a large model to address the difficulties of existing technologies is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for electronic medical record analysis and personalized rehabilitation recommendation based on a large model, which can automatically and accurately analyze electronic medical records and generate personalized rehabilitation recommendation plans based on the analysis results.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for electronic medical record analysis and personalized rehabilitation recommendation based on a large model includes: S1. Collect multi-source medical data of patients from the hospital information system; S2. Preprocess the collected multi-source medical data; S3. Integrate the preprocessed multi-source medical data to construct a multimodal patient profile; S4. Input the constructed patient profile into the pre-trained artificial intelligence analysis model for multi-dimensional analysis and obtain the analysis results; S5. Based on clinical rehabilitation guidelines and expert consensus, a personalized rehabilitation recommendation engine is built; S6. Based on the analysis results, generate personalized rehabilitation recommendation plans through a personalized rehabilitation recommendation engine; S7. During the implementation of the personalized rehabilitation recommendation plan, continuously collect patient feedback data and use the feedback data to incrementally learn and retrain the artificial intelligence analysis model in step S4.

[0007] In the above method, optionally, in S1, the multi-source medical data includes: structured data, unstructured text data, and medical image data; Structured data, including: vital signs and test results; Unstructured text data includes: patient-reported data, admission records, medical records, and discharge summaries.

[0008] Optionally, in the above method, in S2, the collected multi-source medical data is preprocessed, specifically as follows: Natural language processing is performed on unstructured text data, including named entity recognition and relation extraction, extracting key medical entities and their relationships, and transforming them into structured data.

[0009] Optionally, in S3 of the above method, the preprocessed multi-source medical data is fused to construct a multimodal patient profile, specifically as follows: All preprocessed data are standardized and structured, and then fused to generate a multimodal data sequence; Based on the multimodal data sequence, construct a multimodal patient profile; Multimodal patient profiles include: basic profile, disease profile, physiological profile, and treatment profile; Basic profile, including: demographic information, medical history, and allergy history; Disease profile, including: current diagnosis, disease stage / grading, and complications; Physiological profiling includes: physiological indicator trends and organ function status; Treatment profile, including: history and current medications, surgical records, and treatment outcome assessments.

[0010] The above method, optionally, includes the following artificial intelligence analysis model in S4: The disease severity assessment model uses a deep learning model to quantitatively score the patient's overall condition. Complication risk prediction model, based on time series prediction model, predicts the probability of a patient developing a specific complication in the future; The rehabilitation potential and prognosis prediction model is based on machine learning models to assess the patient's recovery speed and the ultimate recovery status that can be achieved.

[0011] Optionally, in S5, a personalized rehabilitation recommendation engine is constructed based on clinical rehabilitation guidelines and expert consensus, as described above: S501. Construct a context encoder and a goal parser to parse the analysis results of the artificial intelligence analysis model; use the context encoder to encode multi-dimensional patient features into a unified vector representation; use the goal parser to decompose rehabilitation goals into progressive and measurable sub-goals based on clinical pathways. S502. Construct a rehabilitation knowledge graph to store structured medical knowledge, clinical rehabilitation guidelines, and expert consensus. S503. Construct a rehabilitation recommendation program template library, store validated standardized rehabilitation recommendation program templates, and label them by disease, stage, and severity. S504. Construct a similar case library to store historical successful rehabilitation cases after anonymization, including: patient profiles, recommended rehabilitation plans implemented, and outcomes.

[0012] Optionally, in S6, based on the analysis results, a personalized rehabilitation recommendation plan is generated through a personalized rehabilitation recommendation engine, as described above: S601. A hybrid search is adopted. Based on the keywords in the analysis results, the baseline protocol template is retrieved from the rehabilitation recommendation protocol template library, and then similar past patient cases are found from the similar case library using vector similarity. S602. Based on the unified vector representation obtained from the context encoder and the sub-targets obtained from the target parser decomposition, a preliminary plan is generated through traceable steps according to the rehabilitation knowledge graph, and a reasoning basis is provided for each step. S603. Based on patient preferences and contraindications, and resource availability, fine-tune the preliminary plan, set adjustment ranges for adjustable parameters in the preliminary plan, and obtain the final plan; S604. Output a personalized rehabilitation recommendation plan that includes a complete final plan, reasoning basis, data traceability, expected goals, and risk warnings.

[0013] Optionally, in step S7 of the above method, during the implementation of the personalized rehabilitation recommendation plan, patient feedback data is continuously collected, and the feedback data is used to incrementally learn and retrain the artificial intelligence analysis model in step S4, specifically as follows: The generated personalized rehabilitation recommendation plan is pushed to medical personnel for review, modification and confirmation; The confirmed treatment plan is then issued to the patient for implementation. We continuously collect data from patients during the treatment process through wearable devices, self-assessment scales, and follow-up interactions. This feedback data, along with doctors' modification records, is used to incrementally learn and retrain the artificial intelligence analysis model.

[0014] A large-model-based electronic medical record analysis and personalized rehabilitation recommendation system, executing any one of the above-described large-model-based electronic medical record analysis and personalized rehabilitation recommendation methods, characterized in that it includes: The data acquisition and preprocessing module, data fusion module, AI intelligent analysis engine, personalized rehabilitation recommendation engine, human-computer interaction and visualization module, and feedback learning and model update module are connected in sequence. The feedback learning and model update module is also connected to the AI ​​intelligent analysis engine; Data acquisition and preprocessing module: used to collect and preprocess multi-source medical data from patients; Data fusion module: used to fuse preprocessed multi-source medical data to construct multimodal patient profiles; AI intelligent analysis engine: Built-in trained disease severity assessment model, complication risk prediction model and rehabilitation potential and prognosis prediction model, used to perform multi-dimensional analysis of patient profiles and obtain analysis results; Personalized rehabilitation recommendation engine: Based on the analysis results, a personalized rehabilitation recommendation plan is generated through the personalized rehabilitation recommendation engine; Human-computer interaction and visualization module: Provides doctors and patients with a graphical interface to display patient profiles, analysis results, rehabilitation plans and implementation progress, and to fine-tune the recommended plans; Feedback learning and model update module: Continuously collects patient feedback data and uses the feedback data to incrementally learn and retrain the artificial intelligence analysis model.

[0015] As can be seen from the above technical solutions, compared with the prior art, the present invention provides a method and system for electronic medical record analysis and personalized rehabilitation recommendations based on a large model, which has the following beneficial effects: it can deeply mine the implicit information in unstructured medical records and construct a comprehensive patient profile; it uses multiple artificial intelligence models to perform multi-dimensional quantitative analysis and prediction, making the assessment results more objective and accurate; the rehabilitation plan is closely integrated with the patient's individual characteristics, risk prediction, and prognosis; it can dynamically adjust the analysis results and rehabilitation recommendations according to the patient's latest data, and continuously optimize the model through a feedback loop, possessing self-evolution capabilities; it provides data-based explanations for rehabilitation recommendations, improving the transparency and clinical acceptability of the plan. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a method for electronic medical record analysis and personalized rehabilitation recommendation based on a large model, provided by this invention; Figure 2 The diagram shows the structure of an electronic medical record analysis and personalized rehabilitation recommendation system based on a large model, as provided by this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0020] Reference Figure 1 As shown, this invention discloses a method for electronic medical record analysis and personalized rehabilitation recommendations based on a large model, including: S1. Collect multi-source medical data of patients from the hospital information system; S2. Preprocess the collected multi-source medical data; S3. Integrate the preprocessed multi-source medical data to construct a multimodal patient profile; S4. Input the constructed patient profile into the pre-trained artificial intelligence analysis model for multi-dimensional analysis and obtain the analysis results; S5. Based on clinical rehabilitation guidelines and expert consensus, a personalized rehabilitation recommendation engine is built; S6. Based on the analysis results, generate personalized rehabilitation recommendation plans through a personalized rehabilitation recommendation engine; S7. During the implementation of the personalized rehabilitation recommendation plan, continuously collect patient feedback data and use the feedback data to incrementally learn and retrain the artificial intelligence analysis model in step S4.

[0021] Furthermore, in S1, multi-source medical data includes: structured data, unstructured text data, and medical image data; Structured data, including: vital signs and test results; Unstructured text data includes: patient-reported data, admission records, medical records, and discharge summaries.

[0022] Furthermore, in S2, the collected multi-source medical data undergoes preprocessing, specifically as follows: Natural language processing is performed on unstructured text data, including named entity recognition and relation extraction, extracting key medical entities and their relationships, and transforming them into structured data.

[0023] Furthermore, in S3, the preprocessed multi-source medical data is fused to construct a multimodal patient profile, specifically as follows: All preprocessed data are standardized and structured, and then fused to generate a multimodal data sequence; Based on the multimodal data sequence, construct a multimodal patient profile; Multimodal patient profiles include, but are not limited to, the following dimensions: basic profile, disease profile, physiological profile, and treatment profile; Basic profile, including: demographic information, medical history, and allergy history; Disease profile, including: current diagnosis, disease stage / grading, and complications; Physiological profiling includes: physiological indicator trends and organ function status; Treatment profile, including: history and current medications, surgical records, and treatment outcome assessments.

[0024] Furthermore, in S4, the artificial intelligence analysis model includes: The disease severity assessment model uses a deep learning model to quantitatively score the patient's overall condition. Complication risk prediction models, based on time-series prediction models (such as LSTM and Transformer), predict the probability of patients developing specific complications (such as infection and thrombosis) in the future; The rehabilitation potential and prognosis prediction model is based on machine learning models (such as gradient boosting trees and random forests) to assess the patient's recovery speed and the final recovery status that can be achieved.

[0025] Furthermore, in S5, a personalized rehabilitation recommendation engine is built based on clinical rehabilitation guidelines and expert consensus, specifically as follows: S501. Construct a context encoder and a goal parser, which are the engine's "eyes and ears," responsible for accurately interpreting input, defining rehabilitation problems, parsing the analysis results of artificial intelligence analysis models (e.g., "left hemiplegia after stroke, NIHSS score of 8, high risk of fall, moderate nutritional risk"), and transforming them into clear rehabilitation decision-making contexts and goals; use a context encoder to encode multi-dimensional patient characteristics (e.g., disease, function, risk, psychosocial) into a unified vector representation; use a goal parser, based on clinical pathways, to decompose rehabilitation goals (e.g., "restore walking ability") into progressive, measurable sub-goals (e.g., "restore walking ability"). S502. Construct a rehabilitation knowledge graph to store structured medical knowledge (disease-rehabilitation measures-expected outcome relationships), clinical guidelines (such as the "Guidelines for Early Rehabilitation Treatment of Stroke in China"), and expert consensus; use a graph database for storage to facilitate associative reasoning (e.g., query "stroke" -> "recovery period" -> "recommended exercise therapy"). S503. Construct a rehabilitation recommendation program template library, store validated standardized rehabilitation recommendation program templates, and label them by disease, stage, and severity. S504. Construct a similar case library to store historical successful rehabilitation cases after anonymization, including: patient profiles, recommended rehabilitation plans implemented, and outcomes.

[0026] Furthermore, in S6, based on the analysis results, a personalized rehabilitation recommendation plan is generated through a personalized rehabilitation recommendation engine, specifically as follows: S601. A hybrid search is adopted. Based on the keywords in the analysis results, the baseline protocol template is retrieved from the rehabilitation recommendation protocol template library, and then similar past patient cases are found from the similar case library using vector similarity. S602. Abandoning "black box" generation, adopting "white box" reasoning chain, based on the unified vector representation obtained by the context encoder and the sub-targets obtained by the target parser decomposition, and generating a preliminary plan through traceable steps according to the rehabilitation knowledge graph, and providing reasoning basis for each step; Specifically, a step-by-step reasoning plan is generated using large models (such as ChatGPT, GPT-4, and Wenxin Yiyan), for example: "Step 1: According to guidelines, the patient should undergo balance training; Step 2: Given the high risk of falls, this must be done under supervision; Step 3: Considering muscle strength, the training intensity should be moderate." The "IF-THEN" logic is executed (e.g., IF fall risk = high risk, THEN environmental requirements = under supervision); precise calculations are performed (e.g., daily protein requirement calculated based on weight and activity level = weight in kg). 1.2 g / kg); Each step of the reasoning plan is "instantiated" using the specific outputs of the rule engine and calculator to form the final solution, and the source of each step is automatically recorded (referenced guide entries, calculation formulas used, and referenced case IDs). S603. Based on patient preferences and contraindications, and resource availability, fine-tune the preliminary plan to ensure that the plan is truly suitable for the individual. Set adjustment ranges for the adjustable parameters in the preliminary plan to obtain the final plan, and present it in a way that is most suitable for the patient to understand and use. Specifically, the plan incorporates the patient's personal preferences ("dislikes swimming"), cultural and dietary habits, and daily routines; considers the patient's home environment (whether there are stairs), availability of rehabilitation equipment, and economic situation; sets adjustment ranges for adjustable parameters in the plan (such as exercise duration and frequency), and allows for adjustments based on subsequent feedback; S604. Output a personalized rehabilitation recommendation plan that includes a complete final plan, reasoning basis, data traceability, expected goals, and risk warnings.

[0027] Furthermore, the personalized rehabilitation recommendation plan generated by the rehabilitation recommendation engine is dynamically adjustable and includes: Setting rehabilitation goals: Based on prognostic predictions, set short-term and long-term quantifiable rehabilitation goals; Recommended rehabilitation content includes: Exercise rehabilitation: Recommend specific types, intensities, frequencies, and durations of exercise; Nutritional diet: Recommended daily calorie intake, nutrient ratio and specific recipes; Medication and follow-up: Remind patients to take medication and recommend personalized follow-up appointment times and examination items; Health education: Pushing articles and videos related to the patient's disease; Recommendation basis and explanation: Provide an explanation based on AI analysis results for each recommendation to enhance the reliability of the solution and the trust between doctors and patients.

[0028] Furthermore, in step S7, during the implementation of the personalized rehabilitation recommendation plan, patient feedback data is continuously collected (such as physiological data collected through wearable devices, patient self-assessment scales, and readmission medical record data). This feedback data is used to incrementally learn and retrain the artificial intelligence analysis model in step S4, specifically as follows: The generated personalized rehabilitation recommendation plan is pushed to medical personnel for review, modification and confirmation; The confirmed treatment plan is then issued to the patient for implementation. We continuously collect data from patients during the treatment process through wearable devices, self-assessment scales, and follow-up interactions. This feedback data, along with doctors' modification records, is used to incrementally learn and retrain the artificial intelligence analysis model.

[0029] Reference Figure 2 As shown, a large-model-based electronic medical record analysis and personalized rehabilitation recommendation system executes any one of the above-described large-model-based electronic medical record analysis and personalized rehabilitation recommendation methods, characterized in that it includes: The data acquisition and preprocessing module, data fusion module, AI intelligent analysis engine, personalized rehabilitation recommendation engine, human-computer interaction and visualization module, and feedback learning and model update module are connected in sequence. The feedback learning and model update module is also connected to the AI ​​intelligent analysis engine; Data acquisition and preprocessing module: used to collect and preprocess multi-source medical data from patients; Data fusion module: used to fuse preprocessed multi-source medical data to construct multimodal patient profiles; AI intelligent analysis engine: Built-in trained disease severity assessment model, complication risk prediction model and rehabilitation potential and prognosis prediction model, used to perform multi-dimensional analysis of patient profiles and obtain analysis results; Personalized rehabilitation recommendation engine: Based on the analysis results, a personalized rehabilitation recommendation plan is generated through the personalized rehabilitation recommendation engine; Human-computer interaction and visualization module: Provides doctors and patients with a graphical interface to display patient profiles, analysis results, rehabilitation plans and implementation progress, and to fine-tune the recommended plans; Feedback learning and model update module: Continuously collects patient feedback data and uses the feedback data to incrementally learn and retrain the artificial intelligence analysis model.

[0030] In one specific embodiment, taking a stroke patient as an example, the hospital admission record, head CT report, blood pressure, blood sugar and other data of a stroke patient are collected; The BERT-BiLSTM-CRF model was used to perform named entity recognition on the "admission record" to extract key information such as "left limb muscle strength grade 3", "slurred speech" and "10-year history of hypertension". The extracted information is integrated with the test data to form a digital profile of the patient. For example, the disease profile is "acute ischemic stroke", the physiological profile includes "weakness of left limb muscles" and "NIHSS score = 8", and the treatment profile includes "currently taking aspirin and atorvastatin". The patient's profile was input into an artificial intelligence analysis model for assessment. The model predicted that the patient's "risk of falling" was high (85% probability), "risk of lung infection" was medium (30% probability), and the probability of "good recovery of self-care ability within 3 months" was 60%. Generate personalized rehabilitation recommendations: Rehabilitation goals: Short-term (within 2 weeks) achievement of sitting up at the bedside; Long-term (3 months) achievement of walking with assistance. Rehabilitation content: Exercise rehabilitation: Given the high risk of falls, bedside balance training and passive movement of the left limbs are recommended under the supervision of a therapist. Early walking is prohibited. Nutritional diet: A low-salt, low-fat diet is recommended, with increased protein intake to prevent infection; Health Education: The system automatically pushes videos on "Fall Prevention Guidelines after Stroke" and "Swallowing Function Training"; Explanation: The interface displays "Bedside balance training is recommended because the system assesses your fall risk as 85% (high risk)"; A week later, the smart bracelet worn by the patient showed an improvement in resting heart rate variability and the patient reported improved muscle strength and sensation. These positive feedback data were collected by the system and used to fine-tune the artificial intelligence analysis model (for example, increasing the probability of "good recovery of self-care ability within 3 months" to 65%) and adjust the intensity of subsequent rehabilitation training accordingly. At the same time, these data were also incorporated into the model training pool to optimize the global model.

[0031] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0032] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for electronic medical record analysis and personalized rehabilitation recommendation based on a large model, characterized in that, include: S1. Collect multi-source medical data of patients from the hospital information system; S2. Preprocess the collected multi-source medical data; S3. Integrate the preprocessed multi-source medical data to construct a multimodal patient profile; S4. Input the constructed patient profile into the pre-trained artificial intelligence analysis model for multi-dimensional analysis and obtain the analysis results; S5. Based on clinical rehabilitation guidelines and expert consensus, a personalized rehabilitation recommendation engine is built; S6. Based on the analysis results, generate personalized rehabilitation recommendation plans through a personalized rehabilitation recommendation engine; S7. During the implementation of the personalized rehabilitation recommendation plan, continuously collect patient feedback data and use the feedback data to incrementally learn and retrain the artificial intelligence analysis model in step S4.

2. The method for electronic medical record analysis and personalized rehabilitation recommendation based on a large model according to claim 1, characterized in that, In S1, multi-source medical data includes: structured data, unstructured text data, and medical image data; Structured data, including: vital signs and test results; Unstructured text data includes: patient-reported data, admission records, medical records, and discharge summaries.

3. The method for electronic medical record analysis and personalized rehabilitation recommendation based on a large model according to claim 2, characterized in that, In S2, the collected multi-source medical data undergoes preprocessing, specifically as follows: Natural language processing is performed on unstructured text data, including named entity recognition and relation extraction, extracting key medical entities and their relationships, and transforming them into structured data.

4. The method for electronic medical record analysis and personalized rehabilitation recommendation based on a large model according to claim 3, characterized in that, In S3, preprocessed multi-source medical data are fused to construct a multimodal patient profile, specifically: All preprocessed data are standardized and structured, and then fused to generate a multimodal data sequence; Based on the multimodal data sequence, construct a multimodal patient profile; Multimodal patient profiles include: basic profile, disease profile, physiological profile, and treatment profile; Basic profile, including: demographic information, medical history, and allergy history; Disease profile, including: current diagnosis, disease stage / grading, and complications; Physiological profiling includes: physiological indicator trends and organ function status; Treatment profile, including: history and current medications, surgical records, and treatment outcome assessments.

5. The method for electronic medical record analysis and personalized rehabilitation recommendation based on a large model according to claim 4, characterized in that, In S4, the artificial intelligence analysis model includes: The disease severity assessment model uses a deep learning model to quantitatively score the patient's overall condition. Complication risk prediction model, based on time series prediction model, predicts the probability of a patient developing a specific complication in the future; The rehabilitation potential and prognosis prediction model is based on machine learning models to assess the patient's recovery speed and the ultimate recovery status that can be achieved.

6. The method for electronic medical record analysis and personalized rehabilitation recommendation based on a large model according to claim 5, characterized in that, In S5, a personalized rehabilitation recommendation engine is built based on clinical rehabilitation guidelines and expert consensus, specifically as follows: S501. Construct a context encoder and a goal parser to parse the analysis results of the artificial intelligence analysis model; use the context encoder to encode multi-dimensional patient features into a unified vector representation; use the goal parser to decompose rehabilitation goals into progressive and measurable sub-goals based on clinical pathways. S502. Construct a rehabilitation knowledge graph to store structured medical knowledge, clinical rehabilitation guidelines, and expert consensus. S503. Construct a rehabilitation recommendation program template library, store validated standardized rehabilitation recommendation program templates, and label them by disease, stage, and severity. S504. Construct a similar case library to store historical successful rehabilitation cases after anonymization, including: patient profiles, recommended rehabilitation plans implemented, and outcomes.

7. The method for electronic medical record analysis and personalized rehabilitation recommendation based on a large model according to claim 6, characterized in that, In S6, based on the analysis results, a personalized rehabilitation recommendation plan is generated through a personalized rehabilitation recommendation engine, specifically as follows: S601. A hybrid search is adopted. Based on the keywords in the analysis results, the baseline protocol template is retrieved from the rehabilitation recommendation protocol template library, and then similar past patient cases are found from the similar case library using vector similarity. S602. Based on the unified vector representation obtained from the context encoder and the sub-targets obtained from the target parser decomposition, a preliminary plan is generated through traceable steps according to the rehabilitation knowledge graph, and a reasoning basis is provided for each step. S603. Based on patient preferences and contraindications, and resource availability, fine-tune the preliminary plan, set adjustment ranges for adjustable parameters in the preliminary plan, and obtain the final plan; S604. Output a personalized rehabilitation recommendation plan that includes a complete final plan, reasoning basis, data traceability, expected goals, and risk warnings.

8. The method for electronic medical record analysis and personalized rehabilitation recommendation based on a large model according to claim 7, characterized in that, S7. During the implementation of the personalized rehabilitation recommendation plan, continuously collect patient feedback data, and use the feedback data to incrementally learn and retrain the artificial intelligence analysis model in step S4, specifically as follows: The generated personalized rehabilitation recommendation plan is pushed to medical personnel for review, modification, and confirmation; The confirmed treatment plan is then implemented by the patient. We continuously collect data from patients during the treatment process through wearable devices, self-assessment scales, and follow-up interactions. This feedback data, along with doctors' modification records, is used to incrementally learn and retrain the artificial intelligence analysis model.

9. A large-model-based electronic medical record analysis and personalized rehabilitation recommendation system, executing the large-model-based electronic medical record analysis and personalized rehabilitation recommendation method as described in any one of claims 1-8, characterized in that, include: The data acquisition and preprocessing module, data fusion module, AI intelligent analysis engine, personalized rehabilitation recommendation engine, human-computer interaction and visualization module, and feedback learning and model update module are connected in sequence. The feedback learning and model update module is also connected to the AI ​​intelligent analysis engine; Data acquisition and preprocessing module: used to collect and preprocess multi-source medical data from patients; Data fusion module: used to fuse preprocessed multi-source medical data to construct multimodal patient profiles; AI intelligent analysis engine: Built-in trained disease severity assessment model, complication risk prediction model and rehabilitation potential and prognosis prediction model, used to perform multi-dimensional analysis of patient profiles and obtain analysis results; Personalized rehabilitation recommendation engine: Based on the analysis results, a personalized rehabilitation recommendation plan is generated through the personalized rehabilitation recommendation engine; Human-computer interaction and visualization module: Provides doctors and patients with a graphical interface to display patient profiles, analysis results, rehabilitation plans and implementation progress, and to fine-tune the recommended plans; Feedback learning and model update module: Continuously collects patient feedback data and uses the feedback data to incrementally learn and retrain the artificial intelligence analysis model.

Images