Artificial intelligence auxiliary medical decision-making system

Abstract

The invention discloses an artificial intelligence auxiliary medical decision-making system, and relates to the technical field of medical assistance. Through the detection suggestion module, preliminary reasoning and observation index judgment can be performed according to patient chief complaint in combination with basic information, and detection suggestions are given in combination with detection cost budget of a patient; through the diagnosis module, preliminary judgment can be made according to detection abnormity, disease classification is judged based on the current diagnosis result, further detection suggestions are given, and finally a doctor makes further diagnosis or treatment suggestions; through the treatment suggestion module, disease causes can be judged according to the diagnosis result, and detection cost budget of a patient is combined, so that the diagnosis and treatment efficiency is improved. Giving out treatment suggestions; the detection suggestion module, the diagnosis module and the treatment suggestion module jointly act to form a medical decision-making system, non-common diseases can be found easily, the problem of cross-department missed diagnosis is solved, the medical budget planning function is achieved, and doctors and patients can provide accurate treatment suggestions easily.

Description

Technical Field

[0001] The present invention relates to the field of medical assistance technology, and in particular to an artificial intelligence-assisted medical decision-making system. Background Art

[0002] Medical decision-making is a systematic, multi-dimensional process that requires striking a balance between scientific assessment and humanistic care, ensuring patient safety and treatment outcomes through a multi-stage process.

[0003] Uncommon diseases are a huge challenge for doctors. For some abnormal symptoms and indicators, it is difficult to think of the cause at the first time. For some diseases, the consequences of delaying the optimal treatment time can be extremely serious. Some symptoms may indicate that the patient has other diseases in the body, but doctors are often not familiar enough with diseases outside their department, resulting in an inability to respond in time and missing the optimal treatment opportunity. A typical case is that a patient has difficulty seeing objects clearly and goes to the ophthalmology department for examination and retinal abnormalities are discovered. The ophthalmology clinic will repair the retina, but it is often difficult to associate the retinal abnormality with the possibility of diabetes. Due to limited medical insurance funds and the different family circumstances of each patient, the treatment options they may choose may have different tendencies. However, doctors themselves are not financial planners, so the treatment recommendations given by AI based on the patient's actual financial ability are extremely valuable. To this end, we propose an artificial intelligence-assisted medical decision-making system. Summary of the Invention

[0004] The purpose of the present invention is to solve the problems mentioned in the above background technology and provide an artificial intelligence assisted medical decision-making system.

[0005] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solutions:

[0006] An artificial intelligence-assisted medical decision-making system, comprising:

[0007] Test recommendation module: Based on the patient's chief complaint and basic information, preliminary reasoning is performed to determine the observation indicators and provide test recommendations based on the patient's test cost budget;

[0008] Diagnosis module: Makes a preliminary judgment on abnormalities detected, determines the disease classification based on the current diagnosis results, and provides further testing recommendations. Ultimately, the doctor makes further diagnosis or treatment recommendations.

[0009] Treatment recommendation module: Determine the cause of the disease based on the diagnosis results and give treatment recommendations based on the patient's testing cost budget.

[0010] Furthermore, the detection recommendation module includes: a data preprocessing module, a core reasoning engine, a detection recommendation generation, a risk warning system and an output generation module; the data preprocessing module includes: structured input processing and knowledge base construction, wherein the structured input processing includes standardization of patient basic information, word segmentation processing of the chief complaint text and structural processing of the medical history; the knowledge base construction includes a disease-symptom knowledge graph based on ICD-11+SNOMEDCT, a decision tree for classification detection items including gold standard detection / differential detection / exclusion detection, and a misdiagnosis risk database formed based on clinical guidelines+expert experience rule base; the core reasoning engine includes multi-dimensional disease matching and dynamic probability calculation, wherein the multi-dimensional disease matching includes symptom matching calculation based on TF-IDF+symptom weight coefficient, risk factor correction based on age correction factor and gender-specific coefficient, and medical history exclusion logic based on drug interference exclusion and confirmed disease exclusion; the dynamic probability calculation includes The Bayesian network construction implemented by PyMC3 captures the temporal feature processing formed by the temporal pattern of medical history based on LSTM, and the probability correction rules formed by the priority of critical values and epidemiological correction; the test recommendation generation includes a hierarchical recommendation system and a cost optimization engine, wherein the hierarchical recommendation system includes core tests with a probability >60% and mandatory recommendations, differential tests with a probability of 10-60%, and exclusion tests with a probability <10% but in line with the differential diagnosis logic; the cost optimization engine includes deduplication of test combinations merged across disease detection items, cost sensitivity grading for basic / intermediate / comprehensive packages, and medical insurance policy mapping connected to the local medical insurance catalog API; the risk warning system includes critical path analysis based on the identification of breakpoints in the test missing → confirmed diagnosis path, a risk quantification model based on Monte Carlo simulation of misdiagnosis probability, and risk level visualization using a medical risk assessment matrix; the output generation module includes intelligent documents, which include natural language generation, multi-version output, and dynamic appendixes.

[0011] Furthermore, the technical implementation path of the detection recommendation module is based on the data processing layer, the computing engine layer and the service interface layer; the data processing layer includes using Apache NiFi to build an ETL pipeline, medical entity recognition using the BioBERT model and the knowledge graph stored in Neo4j; the computing engine layer includes probability calculation using PyTorch Geometric, the rule engine integrating Drools and real-time reasoning using ONNX Runtime; the service interface layer includes a microservice architecture, asynchronous task processing and a cost query interface.

[0012] Furthermore, the diagnosis module includes: a data integration module, a cause analysis engine, a probability calculation layer, an interactive verification module and a report generation system; the data integration module includes: multi-source data fusion, abnormal pattern recognition and interference factor filtering, wherein the multi-source data in the multi-source data fusion includes: test result standardization, time series data alignment and multi-dimensional feature matrix construction; the abnormal patterns in the abnormal pattern recognition include: key indicator slices for implementing three-level abnormality judgment, numerical abnormalities exceeding the reference interval ±3SD, combined abnormalities based on the biochemical indicator linkage mode and time series abnormalities based on gradient changes of continuous detection, and the interference factor filtering is based on the drug impact matrix / detection error probability table; the cause analysis engine includes: direct cause matching and composite cause derivation, the direct cause matching includes a single factor decision tree, path example and exact matching algorithm constructed based on the NCCN guidelines; the composite cause derivation includes a causal graph model based on the GAT network, a time series reasoning engine based on the LSTM network and disease superposition detection; the probability calculation layer includes: a dynamic Bayesian network and composite probability fusion, the dynamic Bayesian network includes a composite probability fusion based on NHANES epidemiological data According to the prior probability injection unit, conditional probability update and uncertainty quantification; the composite probability fusion includes a single cause weight (W1) = direct matching degree × guideline recommendation strength, composite cause weight (W2) = ∑ (abnormal item correlation) × time series consistency, final probability = Sigmoid (αW1+βW2+γ expert experience correction); the interactive verification module includes: a clinical decision loop and a real-time feedback mechanism, wherein the clinical decision loop generates follow-up suggestions based on probability gradient differences; the real-time feedback mechanism automatically updates the knowledge base based on the doctor's correction value; the report generation system The system includes: intelligent document architecture, differential diagnosis appendix and risk warning system; wherein, the intelligent document architecture is based on the diagnosis report, which includes disease portrait, treatment path and prognosis assessment; the differential diagnosis appendix implements graded labeling, including red, yellow and green, wherein red is a critical and severe disease that must be ruled out (probability >5%), yellow is a potential risk that requires follow-up observation, and green is a statistically low-probability event; the risk warning system includes: new disease warning, abnormal item combination does not match the existing knowledge base, time series development violates the conventional pathology model and the response pattern exceeds the machine learning confidence interval.

[0013] Furthermore, the technical implementation path of the diagnostic module is based on a core algorithm stack and real-time reasoning optimization, wherein the core algorithm stack includes a knowledge graph unit based on SNOMEDCT+custom causal graph, a probability calculation unit based on the Pyro probabilistic programming library, an anomaly detection unit based on the IsolationForest integrated algorithm, and a visualization unit based on the Plotly medical-specific template; the real-time reasoning optimization includes building a disease pattern cache pool (LRU algorithm maintains high-frequency paths), implementing GPU-accelerated matrix operations (CUDA implements probabilistic parallel computing), and developing a lightweight model for mobile devices (TensorFlowLite conversion).

[0014] Furthermore, the treatment recommendation module includes: a data fusion module, a plan matching engine, a plan evaluation system, a cost optimization system and a report generation system; wherein, the data fusion module includes multimodal data integration and a treatment plan knowledge base; the plan matching engine includes precise treatment plan derivation and multidimensional constraint processing; the plan evaluation system includes a multidimensional scoring model and dynamic weight allocation; the cost optimization system includes intelligent cost calculation and budget adaptive algorithm; the report generation system includes intelligent document architecture and dynamic decision support.

[0015] Furthermore, the multimodal data integration includes diagnosis result coding conversion, medical history feature extraction and budget dynamic classification; the treatment plan knowledge base includes treatment pathway map, cost relationship database and efficacy evidence library; the precise treatment plan derivation includes guideline-driven matching, similar case recommendation and innovative therapy adaptation; the multidimensional constraint processing includes contraindication filtering, resource availability verification and treatment timing planning; the multidimensional scoring model includes efficacy dimension, safety dimension, economic dimension and convenience dimension; the dynamic weight allocation includes emergency scenario, chronic disease scenario and terminal scenario; the intelligent cost calculation includes basic cost calculation, derivative cost prediction and medical insurance policy mapping; the budget adaptive algorithm includes plan priority sorting, alternative plan recommendation and installment payment planning; the intelligent document architecture includes the main plan and alternative plan matrix, wherein the main plan contains efficacy heat map, risk radar map and cost decomposition tree, and the alternative plan matrix types include radical, balanced and conservative; the dynamic decision support includes real-time sensitivity analysis, doctor-patient decision-making interface and legal document generation.

[0016] Furthermore, the technical implementation path of the treatment recommendation module is based on a core algorithm stack, key technological innovations and a verification system, wherein the core algorithm stack includes a knowledge graph unit based on Apache Jena + OWL reasoning, an optimization calculation unit based on the Gurobi mathematical programming solver, a visualization unit based on the Tableau medical template + D3.js and a real-time calculation unit based on the Flink stream processing engine; the key technological innovations include treatment equivalent identification based on twin network comparison, cost elasticity prediction based on the LSTM timing model and scheme combination optimization based on the quantum annealing algorithm; the verification system includes a three-dimensional verification framework and a continuous optimization mechanism, wherein the three-dimensional verification framework includes clinical rationality formed by a blind evaluation of a multidisciplinary team, economic feasibility formed by verification of the medical insurance actuarial model and legal compliance formed by the medical ethics review module; the continuous optimization mechanism includes a real-world evidence (RWE) feedback channel, a treatment plan failure tracing system and a cost prediction error compensation model.

[0017] The beneficial effects of the present invention are as follows:

[0018] The present invention can make preliminary inferences based on the patient's main complaint and basic information through the test suggestion module, judge the observation indicators, and give test suggestions in combination with the patient's test cost budget; the diagnosis module can make preliminary judgments on test abnormalities, judge the disease classification based on the current diagnosis results, give further test suggestions, and finally the doctor makes further diagnosis or treatment suggestions. The treatment suggestion module can judge the cause of the disease based on the diagnosis results, and give treatment suggestions in combination with the patient's test cost budget; the test suggestion module, the diagnosis module and the treatment suggestion module work together to form a medical decision-making system, which can be conducive to discovering uncommon diseases, solving the problem of missed diagnosis across departments, and also has the function of medical budget planning, which is conducive to providing accurate treatment suggestions for doctors and patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a working schematic diagram of the present invention;

[0020] Figure 2 It is a workflow diagram of the detection suggestion module in the present invention;

[0021] Figure 3 It is a workflow diagram of the diagnosis module in the present invention;

[0022] Figure 4 It is a workflow diagram of the treatment suggestion module in the present invention. DETAILED DESCRIPTION

[0023] To make the objectives, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0024] See also Figure 1 - Figure 4 The present invention provides an artificial intelligence-assisted medical decision-making system, comprising:

[0025] Test recommendation module: Based on the patient's chief complaint and basic information, preliminary reasoning is performed to determine the observation indicators and provide test recommendations based on the patient's test cost budget;

[0026] Diagnosis module: Makes a preliminary judgment on abnormalities detected, determines the disease classification based on the current diagnosis results, and provides further testing recommendations. Ultimately, the doctor makes further diagnosis or treatment recommendations.

[0027] Treatment recommendation module: Determine the cause of the disease based on the diagnosis results and give treatment recommendations based on the patient's testing cost budget.

[0028] In this embodiment, preferably, the detection suggestion module includes: a data preprocessing module, a core reasoning engine, a detection suggestion generation, a risk warning system and an output generation module; the data preprocessing module includes: structured input processing and knowledge base construction, wherein the structured input processing includes standardization of patient basic information, word segmentation processing of the chief complaint text and structural processing of the medical history; the knowledge base construction includes a disease-symptom knowledge graph based on ICD-11+SNOMEDCT, a decision tree for classification detection items including gold standard detection / differential detection / exclusion detection, and a misdiagnosis risk database formed based on clinical guidelines+expert experience rule base; the core reasoning engine includes multi-dimensional disease matching and dynamic probability calculation, wherein the multi-dimensional disease matching includes symptom matching calculation based on TF-IDF+symptom weight coefficient, risk factor correction based on age correction factor and gender-specific coefficient, and medical history exclusion logic based on drug interference exclusion and confirmed disease exclusion; the dynamic probability calculation includes It is built on a Bayesian network implemented in PyMC3, processes temporal features formed by capturing temporal patterns in medical history based on LSTM, and implements probability correction rules formed based on priority of critical values and epidemiological corrections. The generation of test recommendations includes a hierarchical recommendation system and a cost optimization engine. The hierarchical recommendation system includes core tests with a probability >60% that are mandatory, differential tests with a graded recommendation of 10-60% and exclusion tests with a probability <10% that meet the logic of differential diagnosis. The cost optimization engine includes deduplication of test combinations that merge across disease testing items, cost sensitivity grading for basic / intermediate / comprehensive packages, and medical insurance policy mapping that connects to the local medical insurance catalog API. The risk warning system includes critical path analysis based on the identification of breakpoints in the path from missing tests to confirmed diagnosis, a risk quantification model based on Monte Carlo simulation of misdiagnosis probability, and risk level visualization using a medical risk assessment matrix. The output generation module includes intelligent documents, which include natural language generation, multi-version output, and dynamic appendices.

[0029] In this embodiment, preferably, the technical implementation path of the detection suggestion module is based on the data processing layer, the computing engine layer and the service interface layer; the data processing layer includes using Apache NiFi to build an ETL pipeline, medical entity recognition using the BioBERT model and the knowledge graph stored in Neo4j; the computing engine layer includes probability calculation using PyTorch Geometric, the rule engine integrating Drools and real-time reasoning using ONNX Runtime; the service interface layer includes a microservice architecture, asynchronous task processing and a cost query interface.

[0030] In this embodiment, preferably, the diagnosis module includes: a data integration module, a cause analysis engine, a probability calculation layer, an interactive verification module and a report generation system; the data integration module includes: multi-source data fusion, abnormal pattern recognition and interference factor filtering, wherein the multi-source data in the multi-source data fusion includes: test result standardization, time series data alignment and multi-dimensional feature matrix construction; the abnormal patterns in the abnormal pattern recognition include: key indicator slices for implementing three-level abnormality judgment, numerical abnormalities exceeding the reference interval ±3SD, combined abnormalities based on the biochemical indicator linkage mode and time series abnormalities based on the gradient change of continuous detection, and interference factor filtering is based on the drug impact matrix / detection error probability table; the cause analysis engine includes: direct cause matching and composite cause derivation, direct cause matching includes a single factor decision tree, path example and exact matching algorithm constructed based on the NCCN guidelines; composite cause derivation includes a causal graph model based on the GAT network, a time series reasoning engine based on the LSTM network and disease superposition detection; the probability calculation layer includes: dynamic Bayesian network and composite probability fusion, dynamic Bayesian network includes a composite probability fusion based on the NHANES epidemiology The system includes a data prior probability injection unit, conditional probability update, and uncertainty quantification; composite probability fusion includes a single causal weight (W1) = direct match × guideline recommendation strength, a composite causal weight (W2) = ∑(anomaly item correlation) × temporal fit, and a final probability = Sigmoid (αW1 + βW2 + γ expert experience correction); the interactive verification module includes a clinical decision loop and a real-time feedback mechanism, wherein the clinical decision loop generates follow-up suggestions based on probability gradient differences; the real-time feedback mechanism automatically updates the knowledge base based on the physician's correction value; the report generation system includes an intelligent document architecture, a differential diagnosis appendix, and a risk warning system; the intelligent document architecture is based on the confirmed report, which includes a disease profile, treatment pathway, and prognosis assessment; the differential diagnosis appendix implements graded labeling, including red, yellow, and green, where red indicates critical and severe conditions that must be ruled out (probability >5%), yellow indicates potential risks requiring follow-up observation, and green indicates statistically low-probability events; the risk warning system includes warnings for new diseases, abnormal item combinations that do not match the existing knowledge base, temporal development that violates the conventional pathology model, and response patterns that exceed the machine learning confidence interval.

[0031] In this embodiment, preferably, the technical implementation path of the diagnosis module is based on the core algorithm stack and real-time reasoning optimization, wherein the core algorithm stack includes a knowledge graph unit based on SNOMEDCT+custom causal graph, a probability calculation unit based on the Pyro probabilistic programming library, an anomaly detection unit based on the IsolationForest integrated algorithm, and a visualization unit based on the Plotly medical-specific template; real-time reasoning optimization includes building a disease pattern cache pool (LRU algorithm maintains high-frequency paths), implementing GPU-accelerated matrix operations (CUDA implements probabilistic parallel computing), and developing a lightweight model for mobile devices (TensorFlowLite conversion).

[0032] In this embodiment, preferably, the treatment recommendation module includes: a data fusion module, a plan matching engine, a plan evaluation system, a cost optimization system and a report generation system; wherein, the data fusion module includes multimodal data integration and a treatment plan knowledge base; the plan matching engine includes precise treatment plan derivation and multidimensional constraint processing; the plan evaluation system includes a multidimensional scoring model and dynamic weight allocation; the cost optimization system includes intelligent cost calculation and budget adaptive algorithm; the report generation system includes intelligent document architecture and dynamic decision support.

[0033] In this embodiment, preferably, multimodal data integration includes diagnosis result coding conversion, medical history feature extraction and budget dynamic classification; the treatment plan knowledge base includes a treatment pathway map, a cost relationship database and an efficacy evidence base; precise treatment plan derivation includes guideline-driven matching, similar case recommendation and innovative therapy adaptation; multidimensional constraint processing includes contraindication filtering, resource availability verification and treatment timing planning; the multidimensional scoring model includes efficacy dimension, safety dimension, economic dimension and convenience dimension; dynamic weight allocation includes emergency scenarios, chronic disease scenarios and terminal scenarios; intelligent cost calculation includes basic cost calculation, derivative cost prediction and medical insurance policy mapping; the budget adaptive algorithm includes plan priority sorting, alternative plan recommendation and installment payment planning; the intelligent document architecture includes a main plan and an alternative plan matrix, wherein the main plan includes an efficacy heat map, a risk radar map and a cost decomposition tree, and the alternative plan matrix types include radical, balanced and conservative; dynamic decision support includes real-time sensitivity analysis, a doctor-patient decision-making interface and legal document generation.

[0034] In this embodiment, preferably, the technical implementation path of the treatment recommendation module is based on the core algorithm stack, key technological innovations and verification system, wherein the core algorithm stack includes a knowledge graph unit based on Apache Jena + OWL reasoning, an optimization calculation unit based on the Gurobi mathematical programming solver, a visualization unit based on the Tableau medical template + D3.js and a real-time calculation unit based on the Flink stream processing engine; key technological innovations include treatment equivalent identification based on twin network comparison, cost elasticity prediction based on the LSTM timing model and scheme combination optimization based on the quantum annealing algorithm; the verification system includes a three-dimensional verification framework and a continuous optimization mechanism, wherein the three-dimensional verification framework includes clinical rationality formed by a blind evaluation of a multidisciplinary team, economic feasibility formed by verification of the medical insurance actuarial model and legal compliance formed by the medical ethics review module; the continuous optimization mechanism includes a real-world evidence (RWE) feedback channel, a treatment plan failure tracing system and a cost prediction error compensation model.

[0035] The working principle and use process of the present invention:

[0036] Testing recommendation module: Based on the patient's chief complaint and basic information, preliminary reasoning is conducted to determine the observation indicators, and testing recommendations are given based on the patient's testing cost budget.

[0037] The detection recommendation module includes: data preprocessing module, core reasoning engine, detection recommendation generation, risk warning system and output generation module; the data preprocessing module includes: structured input processing and knowledge base construction, among which, structured input processing includes standardization of patient basic information, word segmentation processing of chief complaint text and structural processing of medical history; knowledge base construction includes disease-symptom knowledge graph based on ICD-11+SNOMEDCT, classification detection item decision tree including gold standard detection / differential detection / exclusion detection, and misdiagnosis risk database formed based on clinical guidelines+expert experience rule base; the core reasoning engine includes multi-dimensional disease matching and dynamic probability calculation, among which, multi-dimensional disease matching includes symptom matching calculation based on TF-IDF+symptom weight coefficient, risk factor correction based on age correction factor and gender-specific coefficient, and medical history exclusion logic based on drug interference exclusion and confirmed disease exclusion; dynamic probability calculation includes PyM C3 implements Bayesian network construction, captures temporal feature processing formed by LSTM-based medical history temporal patterns, and implements probability correction rules based on priority of critical values and epidemiological corrections; test recommendation generation includes a hierarchical recommendation system and a cost optimization engine, wherein the hierarchical recommendation system includes core tests with a probability >60% that are mandatory, differential tests with a graded recommendation of 10-60% probability, and exclusion tests with a probability <10% but that meet the differential diagnosis logic; the cost optimization engine includes deduplication of test combinations merged across disease testing items, cost sensitivity grading for basic / intermediate / comprehensive packages, and medical insurance policy mapping connected to the local medical insurance catalog API; the risk warning system includes critical path analysis based on identification of breakpoints in the path from missing detection to confirmed diagnosis, a risk quantification model based on Monte Carlo simulation of misdiagnosis probability, and risk level visualization using a medical risk assessment matrix; the output generation module includes intelligent documents, which include natural language generation, multi-version output, and dynamic appendices.

[0038] The technical implementation path of the detection recommendation module is based on the data processing layer, computing engine layer and service interface layer; the data processing layer includes using Apache NiFi to build the ETL pipeline, medical entity recognition using the BioBERT model and knowledge graph stored in Neo4j; the computing engine layer includes probability calculation using PyTorch Geometric, rule engine integration with Drools and real-time reasoning using ONNX Runtime; the service interface layer includes microservice architecture, asynchronous task processing and cost query interface.

[0039] Diagnostic module: Makes a preliminary judgment on test abnormalities, determines the disease classification based on the current diagnostic results, gives further test recommendations, and ultimately the doctor makes further diagnosis or treatment recommendations.

[0040] The diagnosis module includes: data integration module, cause analysis engine, probability calculation layer, interactive verification module and report generation system; the data integration module includes: multi-source data fusion, abnormal pattern recognition and interference factor filtering, among which the multi-source data in multi-source data fusion includes: test result standardization, time series data alignment and multi-dimensional feature matrix construction; the abnormal patterns in abnormal pattern recognition include: key indicator slices for implementing three-level abnormality judgment, numerical abnormalities exceeding the reference interval ±3SD, combined abnormalities based on biochemical indicator linkage mode and time series abnormalities based on gradient changes of continuous detection, and interference factor filtering is based on drug influence matrix / detection error probability table; the cause analysis engine includes: direct cause matching and compound cause derivation, direct cause matching includes single-factor decision tree, path example and exact matching algorithm built based on NCCN guidelines; compound cause derivation includes causal graph model based on GAT network, time series reasoning engine based on LSTM network and disease superposition detection; the probability calculation layer includes: dynamic Bayesian network and compound probability fusion, dynamic Bayesian network includes prior probability based on NHANES epidemiological data The system includes a rate injection unit, conditional probability update, and uncertainty quantification; composite probability fusion includes a single causal weight (W1) = direct match × guideline recommendation strength, a composite causal weight (W2) = ∑(anomaly item correlation) × temporal fit, and a final probability = Sigmoid (αW1 + βW2 + γ expert experience correction); the interactive verification module includes a clinical decision loop and a real-time feedback mechanism, wherein the clinical decision loop generates follow-up suggestions based on probability gradient differences; the real-time feedback mechanism automatically updates the knowledge base based on physician corrections; the report generation system includes an intelligent document architecture, a differential diagnosis appendix, and a risk warning system; the intelligent document architecture is based on the confirmed report, which includes a disease profile, treatment pathway, and prognosis assessment; the differential diagnosis appendix implements graded labeling, including red, yellow, and green, where red indicates critical and severe conditions that must be ruled out (probability >5%), yellow indicates potential risks requiring follow-up observation, and green indicates statistically low-probability events; the risk warning system includes warnings for new diseases, abnormal item combinations that do not match the existing knowledge base, temporal development that violates conventional pathology models, and response patterns that exceed the machine learning confidence interval.

[0041] The technical implementation path of the diagnosis module is based on the core algorithm stack and real-time reasoning optimization. The core algorithm stack includes a knowledge graph unit based on SNOMEDCT+custom causal graph, a probability calculation unit based on the Pyro probabilistic programming library, an anomaly detection unit based on the IsolationForest integrated algorithm, and a visualization unit based on the Plotly medical-specific template. Real-time reasoning optimization includes building a disease pattern cache pool (LRU algorithm maintains high-frequency paths), implementing GPU-accelerated matrix operations (CUDA implements probabilistic parallel computing), and developing lightweight models for mobile devices (TensorFlowLite conversion).

[0042] Treatment recommendation module: Determine the cause of the disease based on the diagnosis results and give treatment recommendations based on the patient's testing cost budget.

[0043] The treatment recommendation module includes: data fusion module, plan matching engine, plan evaluation system, cost optimization system and report generation system; among them, the data fusion module includes multimodal data integration and treatment plan knowledge base; the plan matching engine includes precise treatment plan derivation and multidimensional constraint processing; the plan evaluation system includes a multidimensional scoring model and dynamic weight allocation; the cost optimization system includes intelligent cost calculation and budget adaptive algorithm; the report generation system includes intelligent document architecture and dynamic decision support.

[0044] Multimodal data integration includes diagnosis result coding conversion, medical history feature extraction and dynamic budget classification; treatment plan knowledge base includes treatment pathway map, cost relationship database and efficacy evidence base; precise treatment plan derivation includes guideline-driven matching, similar case recommendation and innovative therapy adaptation; multidimensional constraint processing includes contraindication filtering, resource availability verification and treatment timing planning; multidimensional scoring model includes efficacy dimension, safety dimension, economic dimension and convenience dimension; dynamic weight allocation includes emergency scenario, chronic disease scenario and terminal scenario; intelligent cost calculation includes basic cost calculation, derivative cost prediction and medical insurance policy mapping; budget adaptive algorithm includes plan priority sorting, alternative plan recommendation and installment payment planning; intelligent document architecture includes main plan and alternative plan matrix, among which the main plan includes efficacy heat map, risk radar map and cost decomposition tree, and the alternative plan matrix types include radical, balanced and conservative; dynamic decision support includes real-time sensitivity analysis, doctor-patient decision-making interface and legal document generation.

[0045] The technical implementation path of the treatment recommendation module is based on the core algorithm stack, key technological innovations and verification system. The core algorithm stack includes a knowledge graph unit based on Apache Jena + OWL reasoning, an optimization calculation unit based on the Gurobi mathematical programming solver, a visualization unit based on the Tableau medical template + D3.js, and a real-time calculation unit based on the Flink stream processing engine. Key technological innovations include treatment equivalent identification based on twin network comparison, cost elasticity prediction based on the LSTM timing model, and scheme combination optimization based on the quantum annealing algorithm. The verification system includes a three-dimensional verification framework and a continuous optimization mechanism. The three-dimensional verification framework includes clinical rationality formed by a blind evaluation of a multidisciplinary team, economic feasibility formed by verification of the medical insurance actuarial model, and legal compliance formed by the medical ethics review module. The continuous optimization mechanism includes a real-world evidence (RWE) feedback channel, a treatment plan failure tracing system, and a cost prediction error compensation model.

[0046] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present 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 present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An artificial intelligence-assisted medical decision-making system, characterized in that: include: Test recommendation module: Based on the patient's chief complaint and basic information, preliminary reasoning is performed to determine the observation indicators and provide test recommendations based on the patient's test cost budget; Diagnosis module: Makes a preliminary judgment on abnormalities detected, determines the disease classification based on the current diagnosis results, and provides further testing recommendations. Ultimately, the doctor makes further diagnosis or treatment recommendations. Treatment recommendation module: Determine the cause of the disease based on the diagnosis results and give treatment recommendations based on the patient's testing cost budget.

2. The artificial intelligence-assisted medical decision-making system according to claim 1, characterized in that: The detection suggestion module includes: a data preprocessing module, a core reasoning engine, a detection suggestion generation, a risk warning system and an output generation module; the data preprocessing module includes: structured input processing and knowledge base construction, wherein the structured input processing includes standardization of patient basic information, word segmentation processing of the main complaint text and structural processing of the medical history; the knowledge base construction includes a disease-symptom knowledge graph based on ICD-11+SNOMEDCT, a decision tree for classification detection items including gold standard detection / differential detection / exclusion detection, and a misdiagnosis risk database formed based on clinical guidelines+expert experience rule base; the core reasoning engine includes multi-dimensional disease matching and dynamic probability calculation, wherein the multi-dimensional disease matching includes symptom matching calculation based on TF-IDF+symptom weight coefficient, risk factor correction based on age correction factor and gender-specific coefficient, and medical history exclusion logic based on drug interference exclusion and confirmed disease exclusion; the dynamic probability calculation includes a Py The Bayesian network construction implemented by MC3 is based on LSTM to capture the temporal feature processing formed by the temporal pattern of medical history, and the probability correction rules formed by the priority of critical values and epidemiological correction; the test recommendation generation includes a hierarchical recommendation system and a cost optimization engine, wherein the hierarchical recommendation system includes core tests that are mandatory with a probability >60%, differential tests that are graded and recommended with a probability of 10-60%, and exclusion tests that have a probability <10% but meet the differential diagnosis logic; the cost optimization engine includes deduplication of test combinations merged across disease detection items, setting cost sensitivity grading for basic / intermediate / comprehensive packages, and medical insurance policy mapping connected to the local medical insurance catalog API; the risk warning system includes critical path analysis based on the identification of breakpoints in the test missing → confirmed diagnosis path, a risk quantification model based on Monte Carlo simulation of misdiagnosis probability, and risk level visualization using a medical risk assessment matrix; the output generation module includes intelligent documents, which include natural language generation, multi-version output, and dynamic appendixes.

3. The artificial intelligence-assisted medical decision-making system according to claim 1, characterized in that: The technical implementation path of the detection suggestion module is based on the data processing layer, the computing engine layer and the service interface layer; the data processing layer includes using Apache NiFi to build an ETL pipeline, medical entity recognition using the BioBERT model and knowledge graphs stored in Neo4j; the computing engine layer includes probability calculation using PyTorch Geometric, rule engine integration with Drools and real-time reasoning using ONNX Runtime; the service interface layer includes microservice architecture, asynchronous task processing and cost query interface.

4. The artificial intelligence-assisted medical decision-making system according to claim 1, characterized in that: The diagnosis module includes: a data integration module, a cause analysis engine, a probability calculation layer, an interactive verification module and a report generation system; the data integration module includes: multi-source data fusion, abnormal pattern recognition and interference factor filtering, wherein the multi-source data in the multi-source data fusion includes: test result standardization, time series data alignment and multi-dimensional feature matrix construction; the abnormal patterns in the abnormal pattern recognition include: key indicator slices for implementing three-level abnormality judgment, numerical abnormalities exceeding the reference interval ±3SD, combined abnormalities based on the biochemical indicator linkage mode and time series abnormalities based on gradient changes of continuous detection, and the interference factor filtering is based on the drug impact matrix / detection error probability table; the cause analysis engine includes: direct cause matching and composite cause derivation, wherein the direct cause matching includes a single factor decision tree, path example and exact matching algorithm constructed based on the NCCN guidelines; the composite cause derivation includes a causal graph model based on the GAT network, a time series reasoning engine based on the LSTM network and disease superposition detection; the probability calculation layer includes: a dynamic Bayesian network and composite probability fusion, wherein the dynamic Bayesian network includes a composite probability fusion based on the NHANES epidemiological data prior The experimental probability injection unit, conditional probability update and uncertainty quantification; the composite probability fusion includes a single cause weight (W1) = direct matching degree × guideline recommendation strength, a composite cause weight (W2) = ∑ (abnormal item correlation) × time series consistency, and a final probability = Sigmoid (αW1+βW2+γ expert experience correction); the interactive verification module includes: a clinical decision loop and a real-time feedback mechanism, wherein the clinical decision loop generates follow-up suggestions based on probability gradient differences; the real-time feedback mechanism automatically updates the knowledge base based on the doctor's correction value; the report generation system It includes: intelligent document architecture, differential diagnosis appendix and risk warning system; wherein, the intelligent document architecture is based on the diagnosis report, which includes disease portrait, treatment path and prognosis assessment; the differential diagnosis appendix implements graded labeling, including red, yellow and green, wherein red is a critical and severe disease that must be ruled out (probability >5%), yellow is a potential risk that requires follow-up observation, and green is a statistically low-probability event; the risk warning system includes: new disease warning, abnormal item combination does not match the existing knowledge base, time series development violates the conventional pathological model and the response pattern exceeds the machine learning confidence interval.

5. The artificial intelligence-assisted medical decision-making system according to claim 1, characterized in that: The technical implementation path of the diagnostic module is based on a core algorithm stack and real-time reasoning optimization, wherein the core algorithm stack includes a knowledge graph unit based on SNOMEDCT+custom causal graph, a probability calculation unit based on the Pyro probabilistic programming library, an anomaly detection unit based on the IsolationForest integrated algorithm, and a visualization unit based on the Plotly medical-specific template; the real-time reasoning optimization includes building a disease pattern cache pool (LRU algorithm maintains high-frequency paths), implementing GPU-accelerated matrix operations (CUDA implements probabilistic parallel computing), and developing a lightweight model for mobile devices (TensorFlowLite conversion).

6. The artificial intelligence-assisted medical decision-making system according to claim 1, characterized in that: The treatment recommendation module includes: a data fusion module, a plan matching engine, a plan evaluation system, a cost optimization system and a report generation system; wherein, the data fusion module includes multimodal data integration and a treatment plan knowledge base; the plan matching engine includes precise treatment plan derivation and multidimensional constraint processing; the plan evaluation system includes a multidimensional scoring model and dynamic weight allocation; the cost optimization system includes intelligent cost calculation and budget adaptive algorithm; the report generation system includes intelligent document architecture and dynamic decision support.

7. The artificial intelligence-assisted medical decision-making system according to claim 6, characterized in that: The multimodal data integration includes diagnosis result coding conversion, medical history feature extraction and budget dynamic classification; the treatment plan knowledge base includes treatment pathway map, cost relationship database and efficacy evidence library; the precise treatment plan derivation includes guideline-driven matching, similar case recommendation and innovative therapy adaptation; the multidimensional constraint processing includes contraindication filtering, resource availability verification and treatment timing planning; the multidimensional scoring model includes efficacy dimension, safety dimension, economic dimension and convenience dimension; the dynamic weight allocation includes emergency scenario, chronic disease scenario and terminal scenario; the intelligent cost calculation includes basic cost calculation, derivative cost prediction and medical insurance policy mapping; the budget adaptive algorithm includes plan priority sorting, alternative plan recommendation and installment payment planning; the intelligent document architecture includes the main plan and alternative plan matrix, wherein the main plan contains efficacy heat map, risk radar map and cost decomposition tree, and the alternative plan matrix types include radical, balanced and conservative; the dynamic decision support includes real-time sensitivity analysis, doctor-patient decision-making interface and legal document generation.

8. The artificial intelligence-assisted medical decision-making system according to claim 1, characterized in that: The technical implementation path of the treatment recommendation module is based on a core algorithm stack, key technological innovations and a verification system, wherein the core algorithm stack includes a knowledge graph unit based on Apache Jena + OWL reasoning, an optimization calculation unit based on the Gurobi mathematical programming solver, a visualization unit based on the Tableau medical template + D3.js and a real-time calculation unit based on the Flink stream processing engine; the key technological innovations include treatment equivalent identification based on twin network comparison, cost elasticity prediction based on the LSTM timing model and scheme combination optimization based on the quantum annealing algorithm; the verification system includes a three-dimensional verification framework and a continuous optimization mechanism, wherein the three-dimensional verification framework includes clinical rationality formed by a blind evaluation of a multidisciplinary team, economic feasibility formed by verification of the medical insurance actuarial model and legal compliance formed by the medical ethics review module; the continuous optimization mechanism includes a real-world evidence (RWE) feedback channel, a treatment plan failure tracing system and a cost prediction error compensation model.

Images