Scoring system and method for tuberculosis based on AI collaborative logic auditing multi-dimensional risk

Abstract

The invention discloses an AI collaborative logic auditing multi-dimensional risk scoring system and method for tuberculosis, and the system integrates an authoritative diagnosis and treatment guide and an artificial intelligence collaborative reasoning technology, and constructs a structured weight scoring matrix and multi-model logic tracking technology dual-core architecture. Through structured scoring of five-dimension indexes and thinking chain auditing of a Gemini 3 series large model, logic contradiction recognition, offset correction and fusion risk score calculation of clinical data are realized, and meanwhile, visual decision logic, agreement processing suggestions and scientific research data closed-loop management functions are generated. According to the method, an expert-level analysis report can be quickly output, the screening sensitivity and the diagnosis and treatment homogenization level are remarkably improved, and the method is suitable for tuberculosis early screening, clinical decision assistance and scientific research data mining of primary and specialized medical institutions and has extremely high clinical application value and scientific research potential.

Description

Technical Field

[0001] This invention relates to the fields of medical artificial intelligence and tuberculosis screening technology, specifically to a scoring system and method for tuberculosis based on AI collaborative logic auditing of multidimensional risks. Background Technology

[0002] Tuberculosis is a serious infectious disease that poses a significant threat to public health worldwide, and a positive virological test is traditionally the gold standard for diagnosis. However, a large number of patients in clinical practice present with "bacterial-negative tuberculosis" (negative virological test but highly suspicious imaging findings), or are in the process of transitioning from latent infection to active tuberculosis. Screening and diagnosing these patients has become a challenge for prevention and control.

[0003] Existing tuberculosis screening tools are mostly based on fixed-threshold scoring systems, which suffer from rigid indicators and a lack of dynamic logical verification. On the one hand, they cannot effectively integrate multi-dimensional clinical data for comprehensive assessment and are insufficient in capturing patients with atypical imaging manifestations. On the other hand, they struggle to identify logical contradictions in clinical data, easily leading to missed or misdiagnosed cases. Furthermore, traditional screening tools lack interpretability in decision-making, and treatment recommendations lack standardization, making it difficult to achieve homogenized diagnosis and treatment between primary and specialized medical institutions, and failing to meet the needs of research data mining. Therefore, there is an urgent need for an intelligent decision-making system that integrates authoritative guidelines and artificial intelligence technology to reconstruct the early tuberculosis screening workflow and improve screening accuracy and diagnostic efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a scoring system and method for multidimensional risk auditing of tuberculosis based on AI collaborative logic.

[0005] To achieve the above objectives, the present invention provides the following technical solution: This system includes a data entry module, a structured weighted scoring module, an AI collaborative logic auditing module, a fusion risk assessment module, a decision output module, and a data management module. These modules work together to achieve multidimensional risk scoring and intelligent decision-making for tuberculosis.

[0006] Data entry module: Supports the collection of structured indicator data, including epidemiological, imaging, etiological, immunological, clinical phenotype and free text clinical notes, ensuring data comprehensiveness and flexibility. The collected data is stored after standardization processing to provide a basis for subsequent scoring and auditing.

[0007] Structured weighted scoring module: Strictly following the WHO and Chinese classification of tuberculosis standards, a five-dimensional weighted model is constructed. Among them, the imaging feature dimension innovatively adopts the Max+0.5 * Others multi-feature overlay algorithm, which breaks through the limitations of single-feature scoring and objectively restores pathological changes; the other dimensions quantify the indicators according to preset weight rules and output the initial guideline score.

[0008] The AI-powered collaborative logic auditing module integrates the Gemini 3 series large-scale language models and employs multi-model logic tracing technology to perform thought chain auditing. On one hand, it identifies logical paradoxes in clinical data through logical contradiction auditing, such as an extremely low BMI with gaps but a negative etiological test, indicating a risk of missed diagnosis. On the other hand, through offset correction, it mines implicit information in free-text clinical notes, such as a persistent cough for over three months without response to anti-inflammatory treatment, dynamically adjusting the initial guideline score to ensure its clinical authenticity.

[0009] The integrated risk assessment module calculates the integrated risk score based on the initial guideline score and the audit results of AI collaborative logic, and classifies patients into six risk levels—none, low, medium, high, very high, and confirmed—according to preset thresholds, providing a basis for subsequent diagnosis and treatment recommendations.

[0010] Decision output module: Transforms static treatment guidelines into dynamic algorithms, automatically generating standardized clinical pathway recommendations based on different risk levels, such as bronchoscopic intervention, experimental treatment, and home follow-up, to achieve homogenization of diagnosis and treatment; at the same time, it outputs a visual audit report, which shows in detail the expert's logical chain of thought and the weight contribution of each clinical feature, taking into account the needs of decision support, scientific research and clinical teaching.

[0011] Data Management Module: Employs encrypted storage and dynamic cloud synchronization technologies to ensure the security and traceability of case data; enables digital tracking of the entire lifecycle of cases, forming a closed loop from initial screening, assessment, collaborative reasoning to diagnosis and record-keeping; supports batch analysis and export of clinical data and secondary AI auditing of historical medical records, providing high-quality data samples for research projects.

[0012] Implementation method The implementation method of the present invention includes the following steps: 1. Data Acquisition and Structured Processing: Clinical data, including structured indicator data, is collected through the data entry module. This includes epidemiological and occupational history, imaging features, pathogen laboratory indicators, immunological indicators, clinical phenotypes, and free-text clinical notes. The data is then standardized, such as by unifying the format and removing outliers, before being stored in an encrypted database.

[0013] 2. Structured Weighted Scoring: The structured weighted scoring module calls a five-dimensional weight model to quantify and score the standardized structured indicator data. Among them, the imaging features dimension uses the "Max + 0.5 * Others" algorithm to superimpose scores on multiple typical features, while the other dimensions are scored according to preset weight rules, such as assigning values ​​to high-risk medical history and quantitative standards for test results, and finally output the initial guideline score.

[0014] 3. AI Collaborative Logic Audit: The AI ​​Collaborative Logic Audit module activates the Gemini 3 series large model to perform thought chain auditing on clinical data. Logical contradiction identification: Traverse clinical data to automatically identify logical paradoxes and generate risk warnings; Offset Correction: Extract implicit and effective information from free-text clinical notes and dynamically adjust the initial guideline scores in a context-aware manner in conjunction with the clinical scenario to correct scoring bias.

[0015] 4. Fusion Risk Assessment: The fusion risk assessment module weights and fuses the initial guideline score and the offset correction result to calculate the fusion risk score. Then, based on the preset risk level thresholds, such as no high risk: 0-10 points, low risk: 11-20 points, etc., the patient's risk level is classified.

[0016] 5. Decision Output: The decision output module matches corresponding standardized clinical pathway recommendations based on the risk level, and generates a visual audit report, which includes the expert's logical thinking chain, the weight contribution of each clinical feature, and the risk level description. The report is presented to clinicians through the terminal, with a total time of no more than 10 seconds.

[0017] 6. Data Management and Scientific Research Applications: The data management module encrypts and stores case data and synchronizes it with the cloud, enabling digital tracking throughout the entire lifecycle. It supports physicians in performing batch analysis and export of clinical data through the "Data Perspective" function, or in conducting secondary AI audits of historical medical records through the "Batch Collaborative Reasoning" function, quickly screening potential subclinical patients and providing data support for scientific research.

[0018] The beneficial effects of this invention are as follows: This invention enhances the ability to capture patients with atypical imaging manifestations through a structured scoring system based on five dimensions and a multi-sign overlay algorithm. The AI-powered collaborative logic auditing function effectively identifies logical inconsistencies in the data and corrects scoring deviations, significantly reducing missed diagnoses and misdiagnoses, particularly suitable for screening septic-negative tuberculosis and subclinical tuberculosis. The entire process, from data entry to outputting the audit report, takes no more than 10 seconds, greatly reducing the diagnostic difficulty for primary care physicians. The visualized audit report displays the complete expert thought process chain, while also providing high-quality materials for research and clinical teaching.

[0019] Based on risk levels, it automatically generates standardized clinical pathway recommendations, transforming static guidelines into dynamic algorithms to ensure consistency in treatment plans across different medical institutions and physicians at different levels. Encryption and cloud synchronization technologies guarantee data security and traceability, while full-lifecycle digital tracking meets clinical management needs. Batch analysis and secondary auditing functions support research data mining. It is suitable for early tuberculosis screening and clinical decision support in various scenarios, including primary healthcare institutions, specialized hospitals, and welfare institutions, demonstrating strong practicality and widespread application value.

[0020] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below. Detailed Implementation

[0021] The specific embodiments of the present invention will be described in further detail below with reference to the examples. These examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0022] Unless otherwise specified, the technical solutions described in this invention are conventional solutions in the field; unless otherwise specified, the materials described are all from commercial channels.

[0023] Example 1: System Deployment and Data Acquisition This system is developed based on the Google Gemini 3 Pro collaboration engine and deployed on a cloud server, supporting access by medical institutions via computer terminals and tablet terminals. Clinicians enter patient information through the data entry module: Epidemiological and occupational history: The patient is HIV-infected, has a history of close contact with a tuberculosis patient, and works as a hospital caregiver; Imaging findings: Chest CT showed tree-in-bud sign and a small number of caseous lesions; Laboratory indicators for pathogens: Xpert test negative, sputum smear negative; Immunological marker: The quantitative value of the QFT test result was 18; Clinical phenotype: Symptoms of tuberculosis intoxication, such as low-grade fever, night sweats, and fatigue; Clinical Notes (Free Text): A patient had been experiencing a recurring cough for over four months, which was unresponsive to anti-inflammatory treatment.

[0024] Example 2: Scoring and Auditing Process S1. Structured Weighted Scoring: The structured weighted scoring module scores according to preset rules. The epidemiology and occupational history dimension gets 25 points, imaging signs get 18 points through the Max + 0.5 * Others algorithm, pathogen laboratory indicators get 5 points, immunological indicators get 20 points, clinical phenotype gets 15 points, and the initial guideline score totals 83 points.

[0025] S2, AI Collaborative Logic Audit: The Gemini 3 Pro model executed a thought chain audit, but failed to identify logical contradictions; the implicit information "repeated cough for more than 4 months, ineffective anti-inflammatory treatment" was extracted from the free text, and the initial guideline score was offset and corrected, resulting in a fusion risk score of 88 points.

[0026] 3. Risk level classification: Based on the preset threshold, the patient is classified as extremely high risk (80-90 points).

[0027] Example 3: Decision Output and Application The decision output module generates a standardized clinical pathway recommendation: "Bronchoscopic interventional examination is recommended to further clarify the diagnosis." It also outputs a visualized audit report, showing the weighted contribution of each indicator and the basis for AI corrections. After reviewing the report, the physician initiates bronchoscopy, ultimately confirming a diagnosis of subclinical tuberculosis, and the patient is admitted to standardized treatment. The data management module encrypts, stores, and tracks the patient's data, and this case can also be exported as a research data sample.

[0028] Example 4: Application of Scientific Data Mining A medical institution needs to conduct a study on the accuracy of tuberculosis screening for HIV-infected individuals. By exporting screening data of 500 HIV-infected individuals through the data management module of this system, and using the batch collaborative reasoning function for secondary AI auditing, 32 potential subclinical patients with initially moderate scores but who were corrected by AI were quickly screened out, providing core data samples for the study and significantly improving research efficiency.

[0029] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0030] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0031] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A scoring system for tuberculosis based on AI collaborative logic auditing of multidimensional risks, characterized in that, It includes a data entry module, a structured weighted scoring module, an AI collaborative logic auditing module, a fusion risk assessment module, a decision output module, and a data management module; The data entry module is used to collect and structure clinical data, including epidemiological and occupational history data, imaging signs data, pathogen laboratory indicator data, immunological indicator data and clinical phenotype data, and also supports free text clinical record entry. The structured weighted scoring module constructs a five-dimensional weighted model, uses the Max+0.5 *Others multi-feature overlay algorithm to score typical imaging features, quantifies and scores the remaining dimensional indicators according to preset weight rules, and outputs the initial guideline score. The AI ​​collaborative logic auditing module integrates the Gemini 3 series large-scale language model to perform thought chain auditing, including logical contradiction auditing and offset correction. The logical contradiction auditing is used to identify logical paradoxes in clinical data. The offset correction is used to combine implicit information in free text clinical notes to make context-aware dynamic adjustments to the initial guideline score. The integrated risk assessment module calculates and generates an integrated risk score based on the initial guideline score and the audit results of the AI ​​collaborative logic, and classifies the risk level according to a preset threshold. The risk level includes: none, low, medium, high, extremely high risk, and confirmed. The decision output module automatically generates standardized clinical pathway recommendations based on the risk level, and simultaneously outputs a visual audit report containing expert thought logic chains and the weighted contributions of each clinical feature. The data management module employs encryption and cloud-based dynamic synchronization technologies to achieve secure storage, traceability, and full lifecycle digital tracking of case data. It also supports batch analysis and export of clinical data and secondary AI auditing of historical medical records.

2. The system according to claim 1, characterized in that, The five-dimensional weighting model specifically includes: Epidemiological and occupational history dimensions: covering high-risk medical history, dynamic contact history, and information on occupations at high risk of tuberculosis; Imaging sign space: Includes typical tuberculosis imaging signs such as cavitation, tree-bud sign, and caseous pneumonia, and objectively reconstructs pathological changes through a multi-sign overlay algorithm; Laboratory indicators for etiology: integrating molecular detection, sputum smear and sputum culture results; Immunological indicators: Dynamic quantification of QFT / PPD test results; Clinical phenotype dimension: Structured recording of tuberculosis intoxication symptoms.

3. The system according to claim 1, characterized in that, The logical contradiction auditing function of the AI ​​collaborative logic auditing module can automatically identify logical paradoxes in clinical data and issue warnings of missed diagnoses to clinicians.

4. A method for auditing multidimensional risks of tuberculosis based on AI collaborative logic, characterized in that, Includes the following steps: S1: Data Acquisition and Structured Processing: Clinical data, including structured indicator data and free text clinical notes, are collected through the data entry module. The collected data is then standardized and stored in the database. S2: Structured Weighted Scoring: The structured weighted scoring module calls a five-dimensional weight model to quantify and score the standardized structured indicator data. Among them, the imaging features are scored using the Max + 0.5 * Others multi-feature overlay algorithm, and the initial guide score is output. S3: AI Collaborative Logic Audit: The AI ​​Collaborative Logic Audit module performs thought chain audit on clinical data through the Gemini 3 series large model. It first identifies logical contradictions in the data and alerts to risks, and then combines the implicit information in the free text clinical notes to dynamically shift and correct the initial guideline score. S4: Fusion Risk Assessment: The fusion risk assessment module calculates the fusion risk score based on the initial guideline score and the offset correction results, and classifies the risk level according to the preset threshold. S5: Decision Output: The decision output module generates standardized clinical pathway recommendations based on the risk level, and at the same time generates a visual audit report containing the expert's logical chain of thought and the weight contribution of each clinical feature, which is then output to the terminal. S6: Data Management and Scientific Research Applications: The data management module encrypts and stores case data and synchronizes it to the cloud, supports full lifecycle digital tracking, and provides batch data analysis, export, and secondary AI auditing of historical medical records, providing high-quality data samples for scientific research.

5. The method according to claim 5, characterized in that, The visual audit report generated in step S5 can show in detail the weight contribution of each clinical feature to the final risk rating, and can be used for clinical decision support, physician research and clinical teaching.

6. The method according to claim 5, characterized in that, The secondary AI audit function for historical medical records in step S6 supports batch collaborative reasoning on hundreds or thousands of historical medical records to quickly screen out potential subclinical tuberculosis patients.

7. The method according to claim 5, characterized in that, The total time from data entry in step S1 to audit report output in step S5 is no more than 10 seconds.