CN-122000055-A - Grading system and method for tuberculosis based on AI collaborative logic audit multidimensional risk

CN122000055ACN 122000055 ACN122000055 ACN 122000055ACN-122000055-A

Abstract

The invention discloses a grading system and a grading method for tuberculosis based on AI collaborative logic audit multidimensional risk, wherein the system fuses an authoritative diagnosis and treatment guide and an artificial intelligence collaborative reasoning technology to construct a double-core framework of a structured weight grading matrix and a multi-model logic tracking technology. Through the structured scoring of the five-dimension index and the thinking chain audit of the Gemini3 series large model, the logic contradiction identification, offset correction and fusion risk score calculation of clinical data are realized, and the visual decision logic, the protocol treatment suggestion and the scientific research data closed-loop management function are generated. The invention can rapidly output expert analysis reports, remarkably improves screening sensitivity and diagnosis and treatment homogeneity level, is suitable for early tuberculosis screening, clinical decision assistance and scientific research data mining of basic level and special medical institutions, and has extremely high clinical application value and scientific research potential.

Inventors

Gong qinghe
Wan Sibei
GUO QING
CHEN XIAOXIA
QIN ZHONGYING
XU SHENJIN

Assignees

上海市崇明区传染病医院
上海健康医学院附属崇明医院(上海交通大学医学院附属新华医院崇明分院、上海市崇明区中心医院)

Dates

Publication Date: 20260508
Application Date: 20260124

Claims (7)

1. The grading system for the tuberculosis based on the AI collaborative logic audit multidimensional risk is characterized by comprising a data input module, a structured weight grading module, an AI collaborative logic audit module, a fusion risk assessment module, a decision output module and a data management module; The data input module is used for collecting and structurally storing clinical data, including epidemiological and occupational history data, imaging sign data, etiology laboratory index data, immunology index data and clinical phenotype data, and supporting free text clinical note input; The structured weight scoring module is used for constructing a five-dimensional weight model, scoring the imaging typical symptoms by adopting a Max+0. Others multi-symptom superposition algorithm, quantitatively scoring the other dimension indexes according to a preset weight rule, and outputting an initial guideline score; The AI collaborative logic audit module integrates a Gemini 3 series large-scale language model, performs thinking chain audit and comprises logic contradiction audit and offset correction, wherein the logic contradiction audit is used for identifying logic paradox in clinical data, and the offset correction is used for combining implicit information in free text clinical notes to dynamically adjust the context awareness of initial guideline scores; The fusion risk assessment module calculates and generates a fusion risk score based on an initial guideline score and an AI collaborative logic audit result, and divides risk grades according to a preset threshold, wherein the risk grades comprise none, low, medium, high, extremely high risk and definite diagnosis; the decision output module automatically generates a protocol clinical path suggestion according to the risk level, and simultaneously outputs a visual audit report containing expert thinking logic chains and contribution of each clinical feature weight; The data management module adopts encryption and cloud dynamic synchronization technology to realize safe storage, tracing and full life cycle digital tracking of case data, and supports batch analysis and export of clinical data and secondary AI audit of historical medical records.
2. The system according to claim 1, wherein the five-dimensional weight model specifically comprises: epidemiology and occupational history dimensions, which cover high-risk medical history, dynamic contact history and tuberculosis high-risk occupational information; imaging symptom space, namely imaging symptom of typical tuberculosis containing cavity, tree bud symptom and cheese pneumonia, and objectively reducing pathological changes through a multi-symptom superposition algorithm; The index dimension of the etiology laboratory is that integrating molecular detection, sputum smear and sputum culture results; the dimension of immunological index is that QFT/PPD detection results are dynamically quantified; Clinical phenotype dimension structural record tuberculosis toxic symptoms.
3. The system of claim 1, wherein the logic contradiction audit function of the AI collaboration logic audit module is capable of automatically identifying clinical data logic paradox and issuing a missed diagnosis risk prompt to a clinician.
4. The method for auditing multidimensional risks based on AI collaborative logic for tuberculosis is characterized by comprising the following steps: S1, data acquisition and structuring, namely acquiring clinical data including structured index data and free text clinical notes through a data input module, standardizing the acquired data and storing the standardized data into a database; S2, structured weight scoring, namely invoking a five-dimensional weight model by a structured weight scoring module, and quantitatively scoring the standardized structured index data, wherein the imaging signs are scored by adopting a Max+0.5 x Others multi-sign superposition algorithm, and outputting initial guide scores; S3, AI collaborative logic audit, wherein the AI collaborative logic audit module carries out thinking chain audit on clinical data through a Gemini 3 series large model, firstly identifies logic contradictions in the data and prompts risks, and then carries out dynamic offset correction on initial guideline scores by combining implicit information in free text clinical notes; S4, fusion risk assessment, wherein the fusion risk assessment module calculates a fusion risk score according to the initial guideline score and the offset correction result, and classifies risk grades according to a preset threshold; S5, decision output, wherein a decision output module generates a protocol clinical path suggestion according to the risk level, and simultaneously generates a visual audit report containing expert thinking logic chains and contribution of each clinical feature weight, and outputs the visual audit report to a terminal; And S6, the data management and scientific research application comprises the steps that the data management module performs encryption storage and cloud synchronization on case data, supports full life cycle digital tracking, simultaneously provides data batch analysis, export and historical medical records secondary AI audit functions, and provides high-quality data samples for scientific research.
5. The method of claim 5, wherein the visual audit report generated in step S5 is capable of detailing the weight contribution of each clinical feature to the final risk classification for clinical decision assistance, physician research and clinical tape teaching.
6. The method of claim 5, wherein the secondary AI audit function of the historical cases in step S6 supports batch collaborative reasoning of hundreds or thousands of historical cases to rapidly screen out patients with potential subclinical tuberculosis.
7. The method of claim 5, wherein the total elapsed time from the data entry of step S1 to the audit report output of step S5 is no more than 10 seconds.

Description

Grading system and method for tuberculosis based on AI collaborative logic audit multidimensional risk Technical Field The invention relates to the technical field of medical artificial intelligence and tuberculosis screening, in particular to a system and a method for grading multidimensional risk of tuberculosis based on AI collaborative logic audit. Background Tuberculosis is an infectious disease which seriously threatens public health safety in the global scope, and etiology positive is a traditional definite gold standard. However, in clinic, a large number of patients show "mycoyin tuberculosis" with negative etiology but highly suspicious imaging, or subclinical tuberculosis in which latent infection is transitive to active tuberculosis, and screening and diagnosis of such patients becomes a difficulty in prevention and control. The traditional tuberculosis screening tool is mostly based on a scoring system with a fixed threshold value, and has the defects of stiff index and lack of dynamic logic verification, namely, on one hand, the traditional tuberculosis screening tool cannot effectively integrate multi-dimensional clinical data to carry out comprehensive evaluation, has insufficient capturing capability on patients with atypical imaging performance, and on the other hand, logic contradiction in the clinical data is difficult to identify, and missed diagnosis or misdiagnosis is easy to cause. Meanwhile, the traditional screening tool lacks decision interpretation, diagnosis and treatment suggestions lack standardization, homogenization diagnosis and treatment of a basic layer and a special medical institution are difficult to realize, and the requirements of scientific research data mining cannot be met. Therefore, an intelligent decision system integrating an authoritative guideline and an artificial intelligence technology is urgently needed, tuberculosis early screening workflow is reconstructed, and screening accuracy and diagnosis and treatment efficiency are improved. Disclosure of Invention The invention aims to provide a grading system and a grading method for tuberculosis based on AI collaborative logic audit multidimensional risk. In order to achieve the above purpose, the present invention provides the following technical solutions: The system comprises a data input module, a structured weight scoring module, an AI collaborative logic auditing module, a fusion risk assessment module, a decision output module and a data management module, wherein the modules work cooperatively to realize multidimensional risk scoring and intelligent decision of tuberculosis. The data input module supports structured index data, including epidemiology, imaging, etiology, immunology, clinical phenotype and collection of free text clinical notes, ensures data comprehensiveness and flexibility, stores collected data after standardized processing, and provides a basis for subsequent scoring and auditing. And the structured weight scoring module strictly follows WHO and Chinese tuberculosis classification standards to construct a five-dimensional weight model. The imaging sign dimension innovation adopts a Max+0.5X Others multi-sign superposition algorithm, breaks through the limitation of single sign scoring, objectively restores pathological changes, quantifies indexes according to a preset weight rule in other dimensions, and outputs initial guideline scoring. And the AI collaborative logic audit module integrates a Gemini 3 series large-scale language model, and the core executes the thinking chain audit by adopting a multi-model logic tracking technology. On one hand, logical paradox in clinical data is identified through logical contradiction audit, such as extremely low BMI and cavitation but etiology negative, and missed diagnosis risk is prompted, and on the other hand, implicit information in free text clinical notes is mined through an offset correction function, such as repeated cough for 3 months Yu Kang, inflammatory treatment is invalid, initial guideline scores are dynamically adjusted, and clinical authenticity of scores is ensured. And the fusion risk assessment module is used for calculating a fusion risk score based on the initial guideline score and the AI collaborative logic audit result, dividing the patients into six risk grades of no, low, medium, high, extremely high risk and definite diagnosis according to a preset threshold value, and providing a basis for follow-up diagnosis and treatment advice. The decision output module converts the static diagnosis and treatment guide into a dynamic algorithm, automatically generates protocol clinical path suggestions, such as bronchoscope intervention, experimental treatment and home follow-up visit, realizes diagnosis and treatment homogenization, simultaneously outputs a visual audit report, and displays the weight contribution of expert thinking logic chains and clinical characteristics in detail, and meets the requirements of decision a