CN-122024984-A - Intelligent generation method and device for kidney disease outpatient medical record driven by large model

CN122024984ACN 122024984 ACN122024984 ACN 122024984ACN-122024984-A

Abstract

The invention relates to a large model driven intelligent generation method and device for kidney disease clinic medical records, which comprises the steps of selecting kidney disease key indexes and collecting historical clinical data, deriving a first control factor for restraining coupling logic among the kidney disease key indexes based on detection values, trend characteristics and fluctuation characteristics of the kidney disease key indexes, generating a second control factor for indicating a high risk coupling path and/or a third control factor for indicating an reasoning tabu route, acquiring past medical history and follow-up clinical data of a current patient, constructing a global disease course framework in a time dimension, and driving a large model to conduct directional reasoning based on the global disease course framework by utilizing the first control factor, the second control factor and the third control factor to generate a structured medical record reference text comprising index comprehensive interpretation and reference diagnosis results. The invention can realize safe, professional and interpretable intelligent kidney disease medical record generation based on a time sequence disease course frame and multi-index coupling logic.

Inventors

CHEN JING
WANG MENGJING
ZHANG FENGMING
NI LI
MAO JIANPING

Assignees

复旦大学附属华山医院

Dates

Publication Date: 20260512
Application Date: 20260402

Claims (9)

1. A large model driven intelligent generation method for kidney disease clinic medical records is characterized by comprising the following steps: selecting kidney disease key indexes and collecting historical clinical data, wherein the kidney disease key indexes comprise at least one of urine protein, eGFR, blood pressure, body fluid, albumin and immune indexes; Deriving a coupling relation matrix between the kidney disease key indexes in each disease course stage based on the detection values, trend characteristics and fluctuation characteristics of the kidney disease key indexes to obtain a first control factor for restraining coupling logic between the kidney disease key indexes, and generating a second control factor for indicating a high-risk coupling path and/or a third control factor for indicating an inference tabu route; Acquiring the past medical history and follow-up clinical data of the current patient, analyzing and acquiring the course stage, index trend and risk gradient, and thus constructing a global course framework in the time dimension; And driving the large model to conduct directional reasoning based on the global course framework by using the first control factor, the second control factor and the third control factor, and generating a structured medical record reference text comprising index comprehensive interpretation and reference diagnosis results.
2. The method of claim 1, wherein elements in the coupling relation matrix are used to characterize a direction of association and a strength of association between the index pairs.
3. The method according to claim 2, wherein the second control factor and the third control factor are generated by introducing rule correction terms extracted from kidney disease guidelines and/or expert knowledge based on the coupling relation matrix, marking high risk coupling paths and/or generating inference tabu routes, respectively.
4. The method for intelligently generating the medical record according to claim 2, wherein the association direction and the association strength between the pair of characterization indexes are calculated by taking the abnormal state and/or the progress event of the current index as dependent variables and the detection value and the trend characteristic of other indexes as independent variables.
5. The method of claim 1, wherein before the step of driving the large model to perform directional reasoning based on the global course framework, the method further comprises the step of setting an auxiliary diagnosis security domain to enable the generated reference diagnosis result to meet clinical security requirements.
6. The method of claim 5, wherein the reference diagnosis results include a diagnosis reference result and a diagnosis likelihood box, and the auxiliary diagnosis security domain determines whether the global course framework includes a complete diagnosis reasoning evidence chain, and if so, generates a diagnosis reference result, and otherwise generates a diagnosis likelihood box and a list of required supplementary checks.
7. The method of claim 1, further comprising the step of setting a follow-up instruction generator to control the large model to generate follow-up advice including follow-up rhythms, indices to be monitored, possible treatment setpoint and risk pre-warning points at the time of autonomous reasoning, before the step of driving the large model to conduct directional reasoning based on the global course framework.
8. The method for intelligently generating an outpatient medical record according to claim 1, further comprising the step of setting an abnormal link interpreter to control the large model to identify abnormal combinations of indexes and provide interpretation advice during autonomous reasoning before the step of driving the large model to conduct directional reasoning based on the global course framework.
9. A large model driven intelligent generation device for kidney disease outpatient medical records, for implementing the method of claims 1-8, comprising: The input module is used for collecting historical clinical data of kidney disease key indexes, wherein the kidney disease key indexes comprise at least one of urine protein, eGFR, blood pressure, body fluid, albumin and immune indexes; The coupling relation construction module is used for deducing a coupling relation matrix among the kidney disease key indexes in each disease course stage based on the detection value, trend characteristic and fluctuation characteristic of the kidney disease key indexes to obtain a first control factor used for restraining coupling logic among the kidney disease key indexes, and generating a second control factor used for indicating a high-risk coupling path and/or a third control factor used for indicating an inference tabu route; The disease course chain construction module is used for acquiring the past medical history and follow-up clinical data of the current patient, analyzing and acquiring the disease course stage, index trend and risk gradient, so as to construct a global disease course framework in the time dimension; And the reasoning module is used for driving the large model to conduct directional reasoning based on the global course framework by using the first control factor, the second control factor and the third control factor, and generating a structured medical record reference text comprising index comprehensive interpretation and reference diagnosis results.

Description

Intelligent generation method and device for kidney disease outpatient medical record driven by large model Technical Field The invention relates to the technical field of generation of outpatient medical records of kidney diseases, in particular to an intelligent generation method and device of outpatient medical records of kidney diseases driven by a large model. Background The kidney disease is used as a chronic disease and has the characteristics of dependence on longitudinal time sequence disease course, multi-index pathological coupling and progressive diagnosis of an evidence chain, and the existing method for generating the medical record of the outpatient service of the kidney disease has certain defects, so that the method cannot adapt to the special demands: 1) The rule-driven template system adopts static logic mapping, and can only fill content based on fixed fields, and can not carry out modeling analysis on index trend of time dimension, so that core course information can not be embodied, and medical records lack of special decision value; 2) The knowledge base auxiliary diagnosis system adopts single-dimension rule matching, does not construct pathological coupling relation among multiple indexes, cannot process complex clinical situations of multiple pathological manifestations, and does not have complete medical record generation capability; 3) The general large model generating tool is essentially statistical language modeling, does not embed time sequence modeling capability and pathology coupling logic of kidney disease specialty, generates content only based on text correlation, and therefore has three core defects that ① cannot identify clinical significance of index trend, ② is easy to generate content violating pathology logic, ③ tends to output deterministic diagnosis when an evidence chain is incomplete, and diagnosis illusion risks exist. Disclosure of Invention The technical problem to be solved by the invention is to provide the large-model-driven intelligent generation method and device for the kidney disease clinic medical record, which can realize safe, professional and interpretable intelligent kidney disease medical record generation based on a time sequence disease course frame and multi-index coupling logic. The technical scheme adopted by the invention for solving the technical problems is to provide a large model driven intelligent generation method of the outpatient medical record of kidney disease, which comprises the following steps: selecting kidney disease key indexes and collecting historical clinical data, wherein the kidney disease key indexes comprise at least one of urine protein, eGFR, blood pressure, body fluid, albumin and immune indexes; Deriving a coupling relation matrix between the kidney disease key indexes in each disease course stage based on the detection values, trend characteristics and fluctuation characteristics of the kidney disease key indexes to obtain a first control factor for restraining coupling logic between the kidney disease key indexes, and generating a second control factor for indicating a high-risk coupling path and/or a third control factor for indicating an inference tabu route; Acquiring the past medical history and follow-up clinical data of the current patient, analyzing and acquiring the course stage, index trend and risk gradient, and thus constructing a global course framework in the time dimension; And driving the large model to conduct directional reasoning based on the global course framework by using the first control factor, the second control factor and the third control factor, and generating a structured medical record reference text comprising index comprehensive interpretation and reference diagnosis results. Further, the elements in the coupling relation matrix are used for representing the association direction and the association strength between the index pairs. Further, the second control factor and the third control factor are generated by introducing rule correction terms extracted from kidney disease guidelines and/or expert knowledge based on the coupling relation matrix, marking high-risk coupling paths and/or generating inference tabu routes respectively. Further, the association direction and the association strength between the pair of characterization indexes are calculated by taking the abnormal state and/or the progress event of the current index as the dependent variable and the detection value and the trend characteristic of other indexes as the independent variable. Further, before the step of performing directional reasoning on the driving large model based on the global course framework, the method further comprises the step of setting an auxiliary diagnosis security domain so that the generated reference diagnosis result meets clinical security requirements. Further, the reference diagnosis result comprises a diagnosis confirming reference result and a diagnosis possibility frame, the auxil