Rare disease virtual case enhanced generation method and system

Abstract

The invention discloses an enhanced generation method and system for a virtual case of a rare disease, and the method comprises the following steps: S1, extracting core features of the rare disease from an authoritative database through an RAG technology, carrying out the standardization, and screening a matched basic case template for subsequent transformation; s2, non-specific features of the basic case are replaced with rare disease typical expressions based on a rule engine, a disease course timeline is reconstructed, and a structured case framework is generated through medical logic verification; and S3, generating a complete case text by utilizing the medical large model iteration, performing error analysis and knowledge base optimization on substandard cases in combination with independent medical model verification, and outputting high-quality teaching cases. According to the method, the accuracy and timeliness of basic medical knowledge are ensured, and a self-optimization closed loop is formed through a large model iteration generation strategy and independent medical model verification based on error analysis and knowledge base optimization.

Description

Technical Field

[0001] This invention relates to the field of case generation, and more particularly to a method and system for enhancing the generation of virtual cases for rare diseases. Background Technology

[0002] The scarcity of rare disease case resources is a long-standing challenge for medical education and clinical research. Traditional case generation methods mainly rely on manual writing by clinicians, which is inefficient and easily limited by personal experience, making it difficult to systematically cover more than a thousand rare diseases. Existing automated case generation technologies are mostly based on template filling or rule engines, which have obvious limitations: First, template cases are mostly designed for common diseases and lack the specific characteristics of rare diseases. In addition, the symptom descriptions of some existing cases are not aligned with authoritative terminology systems, resulting in low clinical credibility of the generated cases. Second, rule systems have difficulty dynamically integrating the latest medical knowledge, such as newly discovered rare disease gene mutations or clinical manifestations, causing case content to lag behind medical progress. Third, generated cases often contain logical contradictions (such as symptom timing violating the natural history of the disease) or unreasonable values ​​(such as laboratory tests exceeding the typical range), requiring a lot of manual correction before they can be used for teaching.

[0003] Currently, there are several technical solutions related to virtual medical record generation, such as:

[0004] 1. Patent CN202411574159.4 proposes a method for developing virtual case studies with question-and-answer interaction for training nursing staff in patient consultation. However, the application scenario of this patent is limited to training nursing staff in communication skills and cannot be applied to cultivating diagnostic thinking in clinicians; at the same time, this patent lacks case generation for rare diseases, making it difficult to achieve targeted teaching related to rare diseases.

[0005] 2. Patent CN202110675150.2 proposes a method for generating natural person medical records based on virtual human medical records. However, this patent lacks the ability to generate medical records for rare diseases, making it difficult to achieve targeted teaching related to rare diseases. Summary of the Invention

[0006] To address the aforementioned issues, this patent focuses on technological innovation in the following aspects:

[0007] This system utilizes a dynamic knowledge extraction and standardization mechanism based on RAG (Retrieval-Augmented Generation). It connects to authoritative databases in real-time via RAG technology to automatically retrieve core features of rare diseases (including symptoms, laboratory indicators, and natural disease course). It innovatively introduces feature weight calculation (based on the frequency of literature evidence), selecting only high-confidence features for case modification to ensure the accuracy and timeliness of basic medical knowledge. Simultaneously, a weekly dynamic update mechanism is established to automatically capture the latest case reports and papers, supplementing new clinical manifestations.

[0008] A large-scale iterative medical logic verification closed loop is established. Through a large-scale iterative generation strategy and independent medical model verification based on error analysis and knowledge base optimization, a self-optimizing closed loop is formed. The purpose of this invention is to provide a method and system for enhancing the generation of virtual case studies for rare diseases, thereby addressing the problems mentioned in the background section.

[0009] To achieve the above-mentioned objectives, one aspect of the present invention provides a method for enhancing the generation of virtual case reports for rare diseases, comprising the following steps:

[0010] Step S1: Extract and standardize the core features of rare diseases from authoritative databases using RAG technology, and select matching basic case templates for subsequent modification;

[0011] Step S2: Based on the rule engine, the non-specific features of the basic case are replaced with typical manifestations of rare diseases, the disease timeline is reconstructed and verified by medical logic, and a structured case framework is generated.

[0012] Step S3 involves iteratively generating complete case texts using a large medical model, and combining this with independent medical model validation to perform error analysis and knowledge base optimization on cases that do not meet the standards, thereby outputting high-quality teaching cases.

[0013] Furthermore, step S1 includes the following steps:

[0014] Step S101: Extract and standardize the core features of rare diseases from authoritative databases using RAG technology. At the same time, establish a dynamic update mechanism to automatically retrieve the latest published rare disease case reports or papers every week to supplement the feature database with new clinical manifestations.

[0015] Next, the captured features undergo medical standardization processing, using a standardized terminology system (e.g., SNOMED CT, a clinical term in the medical systematic nomenclature) to unify symptom naming. Simultaneously, feature weights are calculated based on literature evidence and clinical report frequency, selecting high-confidence rare disease features with weights >0.7 as the benchmark for subsequent replacement. Finally, a structured feature set for rare diseases is output.

[0016] Step S102: Select basic cases from the standardized case database that are consistent with the target rare disease system, quickly filter incompatible cases through a pre-built medical exclusion rule set to eliminate possible conflicts; mark the common disease diagnoses that already exist in the basic cases to ensure that the integrity of the comorbidity logic is not destroyed during subsequent replacement, and output a basic case set suitable for modification.

[0017] Furthermore, step S2 includes the following steps:

[0018] Step S201: Using a pre-established rare disease feature database, replace the non-specific clinical manifestations in the basic cases with typical features of the target rare disease;

[0019] Step S202: Based on the natural history characteristics of rare diseases, adjust the timeline and disease progression of the basic cases, and re-plan the chronological order and rate of symptom onset to conform to the typical development pattern of the disease; based on the rare disease database, set the disease course parameters:

[0020] Acute illness: Symptom progression time ≤ 7 days;

[0021] Chronic diseases: symptoms lasting for ≥6 months

[0022] Step S203: After completing feature replacement and disease course adjustment, perform multi-level medical logic verification.

[0023] Furthermore, in step S201, replaceable fields in the basic case are identified, and common symptoms are mapped to rare disease-specific manifestations based on the retrieved rare disease diagnostic criteria. During the mapping process, routine examinations in the basic case are retained, but key diagnostic indicators for rare diseases are replaced or supplemented. At the same time, comorbidity information in the basic case that does not conflict with rare diseases is retained, and automatically generated medical annotations are added to common diseases that already exist in the case to ensure the rationality of comorbidity characteristics.

[0024] Furthermore, in step S203:

[0025] First, verify the completeness of the essential diagnostic features to ensure that the core clinical manifestations of the rare disease are included;

[0026] Secondly, verify the reasonable range of each test value and eliminate data combinations that do not conform to medical common sense;

[0027] Finally, examine the coherence of the disease progression description to confirm that the sequence of symptom evolution conforms to the pattern of disease development.

[0028] Furthermore, if any validation rule fails, the process terminates and correctable errors are automatically handled; for cases with serious errors that cannot be corrected, output is prohibited.

[0029] Furthermore, step S3 includes the following steps:

[0030] Step S301: Based on the structured case framework constructed in the previous steps, a complete case text conforming to clinical narrative norms is generated using a large medical model (such as the ClinicalBERT model).

[0031] Step S302: After generating the complete case text, the case content is validated using an independent medical large language model (such as GPT-4Clinical).

[0032] Step S303: When a case fails the verification process, the error analysis process is initiated to accurately locate the root cause of the problem and perform targeted optimization.

[0033] Furthermore, in step S301, standardized features, corrected examination values, and reconstructed disease timelines are input into the large model, and text accuracy is ensured through constrained generation instructions. The generation process adopts an iterative optimization strategy: the first round generates a complete medical record (including chief complaint, present illness, physical examination, and diagnosis), the second round strengthens the generation of key paragraphs (such as details of symptom evolution), and then unifies the expression of medical terminology to output a complete text that conforms to the style of real medical records.

[0034] Furthermore, in step S302, the focus is on evaluating whether the generated cases conform to clinical logic and real-world diagnostic and treatment scenarios to ensure their educational application value. For the input cases, a confidence score is automatically assigned. Cases that meet the criteria are directly added to the case database, while those that do not meet the criteria are returned for correction and optimization, and a regeneration process is triggered if necessary.

[0035] Furthermore, the error analysis process in step S303 handles the following two types of errors separately:

[0036] When medical features in a case conflict with the knowledge graph (e.g., a combination of symptoms that has been disproven), it is considered a feature error. The confidence weight of that feature in the knowledge graph is automatically reduced, and it is marked as requiring review by a clinical expert.

[0037] If the problem stems from inappropriate textual expression (such as vague symptom descriptions or disordered time sequences), it is a generation error. In this case, the generation prompt template of the large model is dynamically adjusted, and specific constraints are added.

[0038] Furthermore, for recurring conflicting characteristics, the system automatically retrieves the latest medical literature for verification, prioritizing updates based on high-quality evidence (such as clinical trial data). This process ensures that each case undergoes rigorous quality control, while simultaneously driving continuous improvement of the knowledge base, ultimately outputting high-quality rare disease cases that meet clinical teaching requirements for verification by clinical experts.

[0039] Another aspect of the present invention provides a virtual case enhancement generation system for rare diseases, comprising a feature extraction module, a case modification module, and a case generation module, wherein:

[0040] The feature extraction module is used to extract core features of rare diseases from authoritative databases using RAG technology and standardize them, and to screen matching basic case templates for subsequent modification;

[0041] The case modification module is used to replace the non-specific features of basic cases with typical manifestations of rare diseases based on the rule engine, reconstruct the disease timeline and generate a structured case framework through medical logic verification.

[0042] The case generation module uses a large medical model to iteratively generate complete case texts, and combines independent medical model verification to perform error analysis and knowledge base optimization on cases that do not meet the standards, outputting high-quality teaching cases.

[0043] Compared with existing technologies, this system and method have the following advantages:

[0044] 1. Utilizing RAG technology, the system connects in real-time to authoritative databases to automatically retrieve core characteristics of rare diseases (including symptoms, laboratory indicators, and natural disease course). It innovatively introduces feature weight calculation (based on the frequency of literature evidence), selecting only high-confidence features for case modification to ensure the accuracy and timeliness of basic medical knowledge. Simultaneously, a weekly dynamic update mechanism is established to automatically capture the latest case reports and papers, supplementing new clinical manifestations.

[0045] 2. A self-optimizing closed loop is formed through a large-scale model iterative generation strategy and independent medical model verification based on error analysis and knowledge base optimization.

[0046] 3. The generated cases are validated to retain the flexibility of clinical judgment while effectively identifying logical contradictions or medical errors that may exist in the generated cases, ensuring that the final output cases are both professionally accurate and practical for teaching. Attached Figure Description

[0047] Figure 1 This is a flowchart of a method for enhancing the generation of virtual case studies for rare diseases.

[0048] Figure 2 This is a flowchart illustrating the method for generating virtual cases of rare diseases, using "Wilson's disease" as an example. Detailed Implementation

[0049] 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.

[0050] like Figure 1 The diagram shown is a flowchart of the method of the present invention. An embodiment of the present invention provides a method for enhancing the generation of virtual case reports for rare diseases, with the specific steps as follows:

[0051] Step S1: Extract core features of rare diseases from authoritative databases using RAG technology and standardize them, then select matching basic case templates for subsequent modification.

[0052] Step S101: First, the system connects in real-time to authoritative rare disease databases (e.g., Orphanet) using RAG technology to retrieve core diagnostic features of registered rare diseases, including typical symptoms, key laboratory test indicators, and descriptions of the natural course of the disease. Simultaneously, a dynamic update mechanism is established to automatically retrieve the latest published rare disease case reports or papers weekly, supplementing the feature database with novel clinical manifestations. Next, the retrieved features undergo medical standardization processing, using a standardized terminology system (e.g., SNOMED CT) to unify symptom naming. Furthermore, feature weights are calculated based on literature evidence and clinical report frequency, selecting high-confidence rare disease features with a weight >0.7 as the benchmark for subsequent replacement. Finally, a structured feature set for rare diseases is output.

[0053] Output example:

[0054] {

[0055] Disease: Wilson's disease

[0056] "Essential characteristics": ["Positive KF ring", "Serum ceruloplasmin <50mg / dL"],

[0057] Exclusion trait: ["Serum ferritin > 500 μg / L"]

[0058] }

[0059] Step S102 involves selecting basic cases from the standardized case database that are consistent with the system involved in the target rare disease (e.g., "Wilson's disease" matching the hepatobiliary system case database). Then, incompatible cases are quickly filtered out using a pre-built set of medical exclusion rules to eliminate potential conflicts. The specific rules are as follows:

[0060]

[0061]

[0062] Common disease diagnoses already existing in the basic cases (such as hypertension and diabetes) are marked to ensure that the integrity of the comorbidity logic is not broken during subsequent replacements. The final output is a basic case set suitable for modification.

[0063] Step S2 involves replacing the non-specific features of the basic case with typical manifestations of rare diseases based on the rule engine, reconstructing the disease timeline, and validating it through medical logic to generate a structured case framework. This includes the following steps:

[0064] Step S201 involves replacing non-specific clinical manifestations in basic cases with typical features of the target rare disease using a pre-established rare disease feature database. First, replaceable fields in the basic cases are identified, including key information such as chief complaint, present medical history, and examination results. Based on retrieved rare disease diagnostic criteria, common symptoms are mapped to rare disease-specific manifestations (e.g., replacing "joint pain" with "morning stiffness lasting more than 1 hour accompanied by symmetrical swelling of small joints").

[0065] During the mapping process, routine examinations in the baseline case (such as complete blood count and biochemical indicators) are retained, but key diagnostic indicators for rare diseases are replaced or supplemented (e.g., refining "abnormal liver function" to "serum ceruloplasmin <50mg / dL"). Simultaneously, comorbidity information in the baseline case that does not conflict with rare diseases is retained, and automatically generated medical annotations (e.g., "avoid using glucocorticoids") are added to common diseases already present in the case to ensure the rationality of comorbidity characteristics.

[0066] Step S202: Based on the natural history characteristics of rare diseases, adjust the timeline and disease progression of the basic cases, and re-plan the chronological order and rate of symptom onset to conform to the typical development pattern of the disease. Based on the rare disease database, set the disease course parameters: acute disease: symptom progression time ≤ 7 days; chronic disease: symptom duration ≥ 6 months.

[0067] At the same time, the temporal correlation between laboratory test results and imaging findings is adjusted to ensure that all test indicators match the disease stage and avoid contradictions between the disease course and test results.

[0068] For example: Enter the basic medical history: "Symptoms lasted for 2 weeks"

[0069] After output correction:

[0070] - 0-3 months: Occasional limb weakness

[0071] - 3-6 months: Difficulty swallowing may occur

[0072] - 6-9 months: Wheelchair assistance required

[0073] Step S203: After completing feature replacement and disease course adjustment, perform multi-level medical logic verification. Specifically:

[0074] First, verify the completeness of essential diagnostic features to ensure they include the core clinical manifestations of the rare disease. Second, validate the reasonable range of all test values, eliminating data combinations that do not conform to medical common sense. Finally, review the coherence of the disease progression description to confirm that the sequence of symptom evolution conforms to the disease's developmental patterns. For any detected logical conflicts, automatically mark them and return modification suggestions until a case framework that conforms to clinical reality is generated.

[0075] The specific verification rules are as follows:

[0076] 1. Diagnostic feature integrity verification.

[0077]

[0078] 2. Verification of numerical reasonableness.

[0079]

[0080] 3. Time logic verification.

[0081]

[0082] If any validation rule fails, the process terminates and correctable errors are automatically handled. For cases with serious errors that cannot be corrected, output is prohibited.

[0083] Step S3 involves iteratively generating complete case texts using a large medical model, and then combining this with independent medical model validation. Cases that do not meet the standards are analyzed for errors and the knowledge base is optimized to ultimately output high-quality teaching cases. This includes the following steps:

[0084] Step S301: Based on the structured case framework constructed in the previous steps, a large medical model (such as the ClinicalBERT model) is used to generate a complete case text that conforms to clinical narrative norms. First, standardized features, corrected examination values, and a reconstructed disease timeline are input into the large model, and constrained generation instructions ensure text accuracy. The generation process employs an iterative optimization strategy: the first round generates a complete medical record (including chief complaint, present illness, physical examination, and diagnosis); the second round enhances the generation of key paragraphs (such as details of symptom evolution); and finally, medical terminology is standardized to output a complete text that conforms to the style of a real medical record.

[0085] The generation instructions are as follows:

[0086] Generate natural and fluent clinical records based on provided medical features, and prohibit the addition of unverified symptoms or test results. Strictly adhere to the following data:

[0087] - Chief complaint: {Replaced chief complaint}

[0088] - Key checks: {Revised laboratory / imaging results}

[0089] - Disease course: {Adjusted timeline}

[0090] Maintain standardized terminology (e.g., use 'increased muscle tone' instead of 'muscle stiffness').

[0091] The optimization instructions are as follows:

[0092] "Enhance the description of symptom development details to ensure consistency with the natural history of the {target disease}. For example:"

[0093] - Evolution from {initial symptoms} to {progressive symptoms}

[0094] - Physical examination record of {specific signs}

[0095] Step S302: After generating the complete case text, an independent medical large language model (such as GPT-4Clinical) is used to validate the case content. This step focuses on evaluating whether the generated cases conform to clinical logic and real-world diagnostic and treatment scenarios, ensuring their educational application value. For the input cases, a confidence score is automatically assigned; cases that meet the score are directly added to the case database, while those that do not meet the score are returned for correction and optimization, triggering a regeneration process if necessary. The designed validation instructions are as follows:

[0096] Please conduct a medical rationale review of the following cases, focusing on the following aspects:

[0097] 1. Does the progression of symptoms conform to the typical manifestations of the [target disease]?

[0098] 2. Do the test results support the clinical presentation?

[0099] 3. Are there any logical flaws in the diagnostic reasoning process?

[0100] Please reply in the following format:

[0101] [Reasonability Score]: 0-100 points

[0102] [Main Issues]: (Write "None" if there are no issues)

[0103] [Suggestions for improvement]: (Write "None" if no suggestions are available)

[0104] Output processing:

[0105] A score of ≥90%: Directly enter the case database

[0106] 80% ≤ Score < 90%: Re-verify after automatic correction

[0107] Score <80%: Marked as requiring human intervention

[0108] This verification mechanism retains the flexibility of clinical judgment while effectively identifying logical contradictions or medical errors that may exist in the generated cases, ensuring that the final output cases are both professionally accurate and practical for teaching.

[0109] Step S303: When a case fails the verification process, an error analysis process is initiated to accurately locate the root cause of the problem and perform targeted optimization. First, error routing is determined, including the following two types:

[0110] Feature-based errors: When medical features in a case conflict with the knowledge graph (such as a combination of symptoms that has been disproven), the confidence weight of that feature in the knowledge graph is automatically reduced, and it is marked as requiring review by a clinical expert.

[0111] Generative errors: If the problem stems from inappropriate textual expression (such as vague symptom descriptions or disordered time sequences), the generated prompt template of the large model will be dynamically adjusted, and specific constraints will be added.

[0112] For recurring conflicting features, the system automatically retrieves the latest medical literature for verification, prioritizing updates based on high-quality evidence (such as clinical trial data). This process ensures that each case undergoes rigorous quality control, while simultaneously driving continuous improvement of the knowledge base, ultimately outputting high-quality rare disease cases that meet clinical teaching requirements for review by clinical experts.

[0113] like Figure 2 As shown, an example of generating virtual cases of rare diseases is illustrated using "Wilson's disease".

[0114] Step S1: Feature extraction and case matching RAG retrieval.

[0115] Step S101: Use RAG technology to search the rare disease database Orphanet to obtain the core features of Wilson's disease:

[0116]

[0117] Added a feature from the latest PubMed literature: "sunflower-like cataract" (a new feature added in 2023).

[0118] Step S102, Basic Case Matching:

[0119] Cases with the chief complaint of "abnormal liver function for 3 months" were screened from the hepatobiliary system case database. Cases with "serum ferritin >500μg / L" were excluded (conflicting with Wilson's disease), and cases with comorbid "juvenile diabetes" were retained. The original basic cases are as follows:

[0120]

[0121]

[0122] Step S2, standardize case modification.

[0123] Step S201, Feature replacement stage.

[0124] Chief complaint modification:

[0125] Original: "Fatigue and loss of appetite for 3 months";

[0126] New: "Feeling weak and losing appetite for 3 months, and experiencing difficulty writing in the past week."

[0127] Check item update:

[0128] Added: Ceruloplasmin 38 mg / dL (this test was not performed in the original case);

[0129] Correction: ALT changed from 120 to 300 U / L (more consistent with the manifestation of acute liver damage in Wilson's disease).

[0130] Additional physical examination information:

[0131] Added: "Positive KF ring in cornea, fine tremor in both hands".

[0132] Step S2, the disease course reconstruction stage.

[0133] Timeline adjustment:

[0134] Original description: "Symptoms lasted for 3 months";

[0135] New description:

[0136] - 0-2 months: Progressive weakness;

[0137] - 2-3 months: Jaundice appears;

[0138] - 3 months+: Fine tremors, difficulty writing.

[0139] Step S203, logic verification.

[0140] Validated by rules:

[0141] It contains all the essential features (KF ring + low ceruloplasmin + high urinary copper).

[0142] The urine copper level was set to "150 μg / 24h" (which is within the range for Wilson's disease).

[0143] Retain the comorbid diabetes and add the note: "Glucose fluctuations need to be monitored".

[0144] Step S3: Case generation and verification.

[0145] Step S301: Generate the large model.

[0146] The structured data obtained in step S2 is input into the large-scale medical model Clinical BERT, which outputs natural language case reports:

[0147]

[0148] Step S302, Independent model verification.

[0149] The generated cases were input into an independent large model, such as GPT-4 Clinical, for validation and scoring, and the following issues were identified:

[0150]

[0151] After automatic correction, the model regenerates virtual cases, re-verifies and optimizes them, and finally enters the rare disease case database to await manual verification by clinical experts.

[0152] Final output case (Wilson's disease):

[0153]

[0154]

[0155] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for enhancing the generation of virtual case studies for rare diseases, characterized in that, Includes the following steps: Step S1: Extract and standardize the core features of rare diseases from authoritative databases using RAG technology, and select matching basic case templates for subsequent modification; Step S2: Based on the rule engine, the non-specific features of the basic case are replaced with typical manifestations of rare diseases, the disease timeline is reconstructed and verified by medical logic, and a structured case framework is generated. Step S3 involves iteratively generating complete case texts using a large medical model, and combining this with independent medical model validation to perform error analysis and knowledge base optimization on cases that do not meet the standards, thereby outputting high-quality teaching cases.

2. The method for enhancing virtual case generation for rare diseases according to claim 1, characterized in that, Step S1 includes the following steps: Step S101: Extract core features of rare diseases from authoritative databases using RAG technology and standardize them. Select rare disease features with a weight > 0.7 as the benchmark for subsequent replacement. At the same time, automatically retrieve the latest published rare disease case reports or papers every week to supplement the feature database with new clinical manifestations. Step S102: Select basic cases from the standardized case database that are consistent with the target rare disease system, quickly filter incompatible cases through a pre-built medical exclusion rule set to eliminate possible conflicts; mark the common disease diagnoses that already exist in the basic cases to ensure that the integrity of the comorbidity logic is not destroyed during subsequent replacement, and output a basic case set suitable for modification.

3. The method for enhancing the generation of virtual case reports for rare diseases according to claim 1, characterized in that, Step S2 includes the following steps: Step S201: Using a pre-established rare disease feature database, replace the non-specific clinical manifestations in the basic cases with typical features of the target rare disease; Step S202: Based on the natural history characteristics of rare diseases, adjust the timeline and disease evolution of the basic cases, and re-plan the time sequence and progression rate of symptoms to make them conform to the typical development pattern of the disease. Step S203: After completing feature replacement and disease course adjustment, perform multi-level medical logic verification.

4. The method for enhancing the generation of virtual case reports for rare diseases according to claim 3, characterized in that, In step S201, replaceable fields in the basic case are identified. Based on the retrieved rare disease diagnostic criteria, common symptoms are mapped to rare disease-specific manifestations. During the mapping process, routine examinations in the basic case are retained, but key diagnostic indicators for rare diseases are replaced or supplemented. At the same time, comorbid information in the basic case that does not conflict with rare diseases is retained, and automatically generated medical annotations are added to common diseases that already exist in the case to ensure the rationality of comorbid characteristics.

5. The method for enhancing the generation of virtual case reports for rare diseases according to claim 3, characterized in that, In step S203: First, verify the completeness of the essential diagnostic features to ensure that the core clinical manifestations of the rare disease are included; Secondly, verify the reasonable range of each test value and eliminate data combinations that do not conform to medical common sense; Finally, examine the coherence of the disease progression description to confirm that the sequence of symptom evolution conforms to the pattern of disease development.

6. The method for enhancing virtual case generation for rare diseases according to claim 1, characterized in that, Step S3 includes the following steps: Step S301: Based on the structured case framework constructed in the previous steps, a complete case text conforming to clinical narrative norms is generated using a large medical model; Step S302: After generating the complete case text, the case content is validated using an independent medical large language model; Step S303: When a case fails the verification process, the error analysis process is initiated to accurately locate the root cause of the problem and perform targeted optimization.

7. The method for enhancing virtual case generation for rare diseases according to claim 6, characterized in that, In step S301, the standardized features, corrected examination values, and reconstructed disease timeline are input into the large model, and the accuracy of the text is ensured through constraint generation instructions. The generation process employs an iterative optimization strategy: the first round generates a complete medical record, the second round enhances the generation of key paragraphs, and then the expression of medical terminology is standardized to output a complete text that conforms to the style of a real medical record.

8. The method for enhancing the generation of virtual case reports for rare diseases according to claim 6, characterized in that, Step S302 focuses on evaluating whether the generated cases conform to clinical logic and real diagnosis and treatment scenarios to ensure their teaching application value; for the input cases, confidence scores are automatically marked; cases that meet the criteria are directly entered into the case database, while cases that do not meet the criteria are returned for correction and optimization, and a regeneration process is triggered if necessary.

9. The method for enhancing the generation of virtual case reports for rare diseases according to claim 6, characterized in that, In step S303, the error analysis process handles the following two types of errors separately: When medical features in a case conflict with the knowledge graph, it is considered a feature error. The confidence weight of that feature in the knowledge graph is automatically reduced, and it is marked as requiring review by a clinical expert. If the problem stems from inappropriate text expression, which is a generation error, then the generation prompt template of the large model will be dynamically adjusted, and specific constraints will be added.

10. A virtual case enhancement generation system for rare diseases, characterized in that, It includes a feature extraction module, a case modification module, and a case generation module, among which: The feature extraction module is used to extract core features of rare diseases from authoritative databases using RAG technology and standardize them, and to screen matching basic case templates for subsequent modification; The case modification module is used to replace the non-specific features of basic cases with typical manifestations of rare diseases based on the rule engine, reconstruct the disease timeline and generate a structured case framework through medical logic verification. The case generation module uses a large medical model to iteratively generate complete case texts, and combines independent medical model verification to perform error analysis and knowledge base optimization on cases that do not meet the standards, outputting high-quality teaching cases.

Images