A chronic atrophic gastritis traditional Chinese medicine wisdom man-machine fusion diagnosis and treatment system

Abstract

The application relates to the field of man-machine integration diagnosis and treatment, and discloses a chronic atrophic gastritis traditional Chinese medicine wisdom man-machine integration diagnosis and treatment system, which comprises the following steps: collecting four-examination information, syndrome differentiation results, medication schemes and historical cases of famous traditional Chinese doctors, forming a standardized special disease diagnosis and treatment data set; analyzing diagnosis and treatment texts, constructing the correlation between symptoms, syndrome differentiation and medication, forming an AI famous doctor subsystem which can automatically analyze the four-examination information and output a preliminary diagnosis and treatment scheme; introducing an inflammation-cancer transformation risk level as a constraint parameter to adjust the diagnosis and treatment analysis process, generate a preliminary syndrome differentiation result and a medication scheme, and generate a scheme proofreading log; iteratively training a syndrome differentiation prediction model and a medication recommendation model, updating model weights and parameters, and generating an updated AI famous doctor subsystem; and analyzing patient diagnosis and treatment information in the special disease database to form a unified patient diagnosis and treatment and risk data set. The application has the advantages of improving diagnosis and treatment efficiency and standardization level.

Description

Technical Field

[0001] This invention relates to the field of human-machine fusion diagnosis and treatment, specifically a traditional Chinese medicine-based intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis. Background Technology

[0002] Traditional Chinese medicine (TCM) possesses unique advantages within China's healthcare system, but its inheritance and application face multiple challenges. Firstly, there is a shortage of talent and insufficient innovation, making it difficult to integrate traditional TCM diagnostic and treatment models with the modern healthcare system, resulting in low or even nonexistent standardization of electronic medical records at the grassroots level. Secondly, the inheritance of renowned veteran TCM doctors and academic schools of thought faces difficulties, posing a risk of cultural discontinuity in TCM and affecting the accumulation and continuation of experience. There is also a lack of intelligent diagnostic and treatment systems specifically designed for chronic atrophic gastritis that integrate the experience of renowned TCM doctors. Therefore, it is essential to design an intelligent human-machine integrated diagnostic and treatment system for chronic atrophic gastritis that improves diagnostic efficiency and standardization. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a TCM-based intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis, which has the advantages of improving diagnostic and treatment efficiency and standardization, and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goals of improving diagnostic and treatment efficiency and standardization, this invention provides the following technical solution: a TCM-based intelligent human-machine fusion diagnostic and treatment system for chronic atrophic gastritis, comprising: Specialized disease database module: Collects information on the four diagnostic methods, syndrome differentiation results, medication plans and historical cases of famous TCM doctors, and records the corresponding data source and time information. The diagnosis and treatment information is uniformly collected and stored in the chronic atrophic gastritis specialized disease database to form a standardized specialized disease diagnosis and treatment dataset. Association Modeling Module: Based on standardized disease diagnosis and treatment datasets, the module parses diagnosis and treatment texts, constructs the association between symptoms, syndrome differentiation, and medication, and trains syndrome differentiation prediction and medication recommendation algorithms to form an AI expert doctor subsystem that can automatically analyze the four diagnostic methods and output preliminary diagnosis and treatment plans. Risk Constraint Module: Input the patient's four diagnostic methods information into the AI ​​Famous Doctor Subsystem, introduce the risk level of inflammation-cancer transformation as a constraint parameter, adjust the diagnosis and treatment analysis process, generate preliminary diagnosis results and medication plans, submit the initial diagnosis plan to the famous doctor inheritance team for proofreading, and generate a plan proofreading log; Model update module: Based on the scheme verification log and newly added case data, the syndrome differentiation prediction model and medication recommendation model are iteratively updated to generate the updated AI expert doctor subsystem; Risk Update Module: Based on the updated AI Expert Doctor Subsystem, the module analyzes patient diagnosis and treatment information in the disease database, updates the corresponding inflammation and cancer transformation risk assessment level, and links and stores the patient diagnosis and treatment data with the updated inflammation and cancer transformation risk level to form a unified patient diagnosis and treatment and risk dataset.

[0005] Preferably, the process of forming a standardized disease-specific diagnosis and treatment dataset is as follows: The collected information from the four diagnostic methods is structured and converted into a unified coding format, including tongue and pulse information, pulse characteristics, symptom descriptions, and lifestyle habits. The historical diagnostic results and medication plans are categorized and standardized, and the diagnosis and treatment records of different famous doctors are mapped into unified semantic tags; Each record is labeled with its source, timestamp, and basic patient information. The structured information from the four diagnostic methods, the results of syndrome differentiation, the medication plan, and historical cases are uniformly stored in the disease-specific database according to the patient dimension, forming a standardized disease-specific diagnosis and treatment dataset.

[0006] The preferred process for establishing the correlation between symptoms, syndrome differentiation, and medication is as follows: Based on a standardized disease diagnosis and treatment dataset, natural language processing is performed on the diagnosis and treatment text to parse symptom descriptions, diagnostic terms, and medication records into computable feature vectors. A multidimensional association matrix is ​​constructed, and symptom characteristics, syndrome differentiation categories and medication labels are encoded accordingly, forming an association weight matrix; Semantic alignment is performed on text data from different sources to unify terminology differences and expression styles, and a standardized symptom-diagnosis-medication mapping relationship is generated.

[0007] Preferably, the process of forming an AI-powered expert doctor subsystem capable of automatically analyzing information from the four diagnostic methods and outputting preliminary treatment plans is as follows: The symptom-syndrome differentiation-medication correlation matrix is ​​vectorized to form structured features that can be input into the model; A multi-task prediction model was constructed, including a syndrome differentiation prediction branch and a medication recommendation branch, while the patient's inflammation-cancer transformation risk level was used as the model input for constraint adjustment; During model training, patient samples of different risk levels are weighted to optimize the model's ability to diagnose and predict medication on high-risk and low-risk samples. The trained model is then packaged into an AI-powered expert doctor subsystem.

[0008] Preferably, the process of generating preliminary diagnosis results and medication plans is as follows: The patient's four diagnostic methods information is vectorized and encoded by the input processing layer of the AI ​​Famous Doctor subsystem, and then combined with the patient's individual characteristics and historical medical data to generate an input feature vector. The input feature vector is fed into the multi-task dialectical prediction branch and the medication recommendation branch, while the patient's inflammation-cancer transformation risk level is introduced as a constraint parameter. The output results are fused, combining the probability of syndrome differentiation and the probability of medication recommendation predicted by each branch, eliminating abnormal predictions that are inconsistent with risk constraints, and generating preliminary syndrome differentiation results and medication plans.

[0009] Preferably, the process of generating the scheme verification log is as follows: The preliminary diagnosis results and medication plan are submitted to the renowned doctors' inheritance team. The expert team reviews the preliminary diagnosis category and medication combination in the visual interface and adjusts the plan by selecting, modifying and adjusting it. For each modification made during the expert review process, record the modification rules, logical basis, and related symptom-diagnosis-medication characteristics; The optimized solution and the corresponding expert knowledge enhancement matrix are stored in the disease-specific database, and a solution verification log is generated synchronously.

[0010] Preferably, the process of generating the updated AI expert doctor subsystem is as follows: Based on the solution verification log, the expert modification operations, adjustment logic, and risk constraint application are extracted and transformed into knowledge-enhanced features. The newly added case data is standardized and characterized, and then merged with knowledge-enhanced features to form an iterative training dataset; Based on the iterative training dataset, a multi-task learning strategy is used to update the parameters of the syndrome differentiation prediction model and the medication recommendation model, while maintaining the risk level constraint adjustment mechanism, to generate the updated AI expert doctor subsystem.

[0011] Preferably, the process for updating the corresponding inflammation-cancer transformation risk assessment level is as follows: Feature extraction and structured coding were performed on patient information from the four diagnostic methods, historical diagnostic results, and medication regimens in the disease-specific database; The extracted patient characteristics are input into the updated AI expert subsystem, and the probability of each patient's inflammation-cancer transformation is calculated by combining the syndrome differentiation prediction branch and the medication recommendation branch with the risk constraint mechanism. Based on the calculated cancer transformation probability and the preset risk level mapping rules, the patient's risk level is updated, and a traceable level modification record is generated.

[0012] The preferred process for forming a unified patient diagnosis and risk dataset is as follows: Obtain the updated patient's risk level for cancer transformation, along with corresponding information from the four diagnostic methods, syndrome differentiation results, and medication regimen; By utilizing the traceable risk level modification records generated during the update process, patient treatment data and historical risk level change information are integrated to form a complete patient record; The integrated records are stored in a disease-specific database to form a unified patient diagnosis and treatment and risk dataset.

[0013] Compared with existing technologies, this invention provides a TCM-based intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis, which has the following beneficial effects: This invention achieves standardized collection and unified management of renowned TCM doctors' diagnostic and treatment data through a disease-specific database module, ensuring the integrity and scalability of diagnostic and treatment information. The association modeling module can parse diagnostic and treatment texts and establish a symptom-syndrome differentiation-medication association model, supporting the AI-based renowned doctor subsystem in automatically generating preliminary treatment plans, improving the efficiency and accuracy of diagnostic and treatment analysis. The risk constraint module introduces the inflammation-cancer transformation risk level as a constraint parameter, and combines it with expert verification and generated plan verification logs to achieve dynamic adjustment of diagnostic and treatment decisions and a traceable feedback mechanism. The model update module uses verification logs and new cases to iteratively update the syndrome differentiation prediction and medication recommendation algorithms, continuously optimizing the AI-based renowned doctor subsystem and improving its diagnostic and treatment reasoning capabilities. The risk update module analyzes patient diagnostic and treatment information in the disease-specific database, updates the inflammation-cancer transformation risk assessment level in real time, and stores it in association with patient data, forming a unified diagnostic and treatment and risk dataset. This enables dynamic linkage between patient information, risk status, and the AI ​​system, thereby comprehensively supporting precise diagnosis and treatment, risk management, and intelligent continuous optimization of chronic atrophic gastritis. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1: Please refer to Figure 1 As shown in the embodiment of the present invention, a TCM intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis includes: Specialized disease database module: Collects information on the four diagnostic methods, syndrome differentiation results, medication plans and historical cases of famous TCM doctors, and records the corresponding data source and time information. The diagnosis and treatment information is uniformly collected and stored in the chronic atrophic gastritis specialized disease database to form a standardized specialized disease diagnosis and treatment dataset.

[0017] The process of creating a standardized disease diagnosis and treatment dataset in the disease-specific database construction module is as follows: The collected diagnostic information was structured, converting tongue and pulse, pulse characteristics, symptom descriptions, and lifestyle habits into a unified coding format. After obtaining the original records from the disease-specific database, a standard coding table was established for tongue characteristics such as tongue body, tongue coating, and tongue shape, mapping textual descriptions such as "pale red" and "obvious teeth marks" into category or numerical codes to make them calculable and easy to analyze. For pulse characteristics such as pulse rate, pulse shape, and pulse strength, a unified pulse coding standard was adopted, converting textual records such as "wiry pulse" and "slippery pulse" into standard category labels. For symptom description fields, natural language processing technology was used to extract and classify keywords, mapping different expressions such as "abdominal distension" and "stomach distension" into unified codes. For lifestyle habits, including diet, rest, and exercise, text or options were converted into fixed format fields through structured questionnaires or predefined options. After processing, each piece of diagnostic information formed a structured record containing tongue characteristics, pulse characteristics, symptom codes, and lifestyle habit codes, providing directly calculable input for data integration and analysis.

[0018] Historical diagnostic results and medication regimens are categorized and standardized, mapping the diagnostic records of different renowned doctors to unified semantic tags. A unified syndrome classification system is established for diagnostic results, mapping syndrome information described by different doctors, such as "liver qi stagnation" and "liver qi invading the stomach," to standard syndrome tags. For medication regimens, drug names, dosages, and usages are coded, a unified drug numbering system is established, and structured tables are formed. For similar diagnostic records provided by multiple doctors, semantic alignment or mapping algorithms are used to unify different expressions to the same category tag, ensuring data consistency. Each historical record ultimately forms computable data containing standard diagnostic tags and structured medication codes, facilitating the establishment of cross-doctor and cross-source diagnostic rules and analysis models.

[0019] Each record is labeled with its source, timestamp, and basic patient information. The structured information from the four diagnostic methods (inspection, auscultation and olfaction, diagnosis), the results of syndrome differentiation, the medication regimen, and historical cases are uniformly stored in a disease-specific database according to the patient dimension, forming a standardized disease-specific diagnosis and treatment dataset. A "data source ID" is added to each record to identify the doctor who provided the record or the data collection channel to ensure data traceability. The record collection timestamp includes the specific date and time, supporting time-series analysis and model training. Basic patient information fields, such as age, gender, disease course, and past medical history, are added, and anonymization or encoding is used to ensure privacy and security. The information from the four diagnostic methods, the results of syndrome differentiation, the medication regimen, and historical cases are integrated into a complete record entry for each patient and stored in a relational database indexed by patient ID to achieve data retrieval and fast access. After complete storage and processing, a sustainable and standardized disease-specific diagnosis and treatment dataset is formed, providing an executable data foundation for symptom-syndrome differentiation-medication correlation analysis, model training, and intelligent diagnosis and treatment systems.

[0020] Association Modeling Module: Based on standardized disease diagnosis and treatment datasets, the module parses diagnosis and treatment texts, constructs the association between symptoms, syndrome differentiation, and medication, and trains syndrome differentiation prediction and medication recommendation algorithms to form an AI expert doctor subsystem that can automatically analyze the four diagnostic methods and output preliminary treatment plans.

[0021] The process of constructing the association between symptoms, syndrome differentiation, and medication in the association modeling module is as follows: Based on standardized disease-specific diagnosis and treatment datasets, natural language processing is performed on the diagnosis and treatment texts to parse symptom descriptions, diagnostic terms, and medication records into computable feature vectors. Structured diagnostic information, standard diagnostic labels, and standardized medication records are obtained from the disease-specific database to form a text input sequence. For the symptom description field, word segmentation algorithms and word embedding methods are used to convert the textual information recorded by different doctors into fixed-dimensional numerical vectors to ensure that the features can be used for calculation and analysis. For the diagnostic terms field, different expressions are uniformly encoded into category vectors by establishing a syndrome type dictionary and category mapping. For the medication record field, including drug codes, dosage, and usage information, multi-dimensional numerical or category vectorization processing is performed so that each record ultimately forms a multi-dimensional feature vector of "symptom vector + diagnostic vector + medication vector", providing directly computable input for constructing the symptom-diagnosis-medication association matrix.

[0022] A multidimensional association matrix is ​​constructed, which encodes symptom features, syndrome differentiation categories, and medication labels to form an association weight matrix. All symptom vectors, syndrome differentiation vectors, and medication vectors are arranged according to patient records to form a three-dimensional data tensor, corresponding to the three dimensions of "symptom-syndrome differentiation-medication". Statistical methods such as co-occurrence frequency calculation, mutual information analysis, or vector similarity-based algorithms are used to quantify the association strength between symptoms and syndrome differentiation, and between syndrome differentiation and medication, generating numerical association weights. The association weight matrix records the numerical weights of each symptom-syndrome differentiation and syndrome differentiation-medication pair, reflecting their statistical dependence and the strength of the correspondence. The matrix is ​​saved in the form of a structured file or database table for direct use in subsequent diagnosis and treatment rule construction or model training.

[0023] Semantic alignment is performed on text data from different sources to unify terminology differences and expression styles, generating standardized symptom-diagnosis-medication mapping relationships. Through word embedding similarity calculation or thesaurus, symptoms with the same meaning but different expressions, such as "abdominal distension" and "stomach distension," as well as diagnosis terms or drug names, from different doctors or historical records are mapped to unified codes. Combined with the association weight matrix, a standardized mapping relationship table is formed. Each record contains a standard symptom code, a standard diagnosis code, a standard medication code, and corresponding association weights. This mapping relationship can be directly used for intelligent diagnosis and treatment systems, data analysis, or model training to realize executable diagnosis and treatment rules across doctors and sources, ensuring a close correlation between structured features and diagnosis and treatment decisions.

[0024] The process of forming an AI-powered expert doctor subsystem in the correlation modeling module, which can automatically analyze information from the four diagnostic methods and output preliminary treatment plans, is as follows: The symptom-syndrome differentiation-medication association matrix is ​​vectorized to form a structured feature that can be input into the model. A standardized symptom-syndrome differentiation-medication association matrix is ​​obtained, which records the numerical weights of each pair of symptoms-syndrome differentiation and syndrome differentiation-medication, reflecting their statistical dependencies. The matrix is ​​converted into a multi-dimensional numerical vector, with each patient record corresponding to a vector containing symptom code, syndrome differentiation code, medication code, and association weight information. During the vectorization process, sparse coding or embedding methods can be combined to compress high-dimensional features into structured features suitable for input to deep learning models, while preserving the associations between features. This enables the model to use the statistical relationship between symptoms, syndrome differentiation, and medication to make syndrome differentiation predictions and medication recommendations.

[0025] A multi-task prediction model is constructed, including a syndrome differentiation prediction branch and a medication recommendation branch. The patient's risk level of inflammation-to-cancer conversion is used as input to constrain and adjust the model. Structured features are used as input, and a multi-task deep neural network is designed. The syndrome differentiation prediction branch outputs the standard syndrome category and is trained using a classification loss function. The medication recommendation branch outputs drug codes and dosage information and is trained using a regression or multi-label classification loss function. Simultaneously, the patient's risk level of inflammation-to-cancer conversion is encoded as an additional input feature and embedded in the network to constrain and adjust the model output. This allows the model to place greater emphasis on key syndrome differentiation features and medication combinations for high-risk patients, thereby achieving differentiated prediction capabilities for patients with different risk levels.

[0026] During model training, patient samples of different risk levels are weighted to optimize the model's diagnostic and medication prediction capabilities on high-risk and low-risk samples. The trained model is then encapsulated as an AI-powered expert doctor subsystem. Different weights are assigned to training samples based on the patient's risk level of cancer-related inflammation, increasing the proportion of high-risk patient samples in the loss function. This improves the model's diagnostic accuracy and medication recommendation rationality for key high-risk patients. Batch gradient descent, forward propagation, backpropagation, and weight update mechanisms are used to complete model training, dynamically adjusting the learning rate to adapt to the sample distribution. After training, the model weights, prediction interface, and input / output processing modules are encapsulated as a callable AI-powered expert doctor subsystem. This subsystem supports automatically generating preliminary diagnostic results and medication plans after inputting the patient's four diagnostic methods information vector, and outputting prediction confidence scores, providing actionable reference data for clinical decision-making.

[0027] Risk Constraint Module: Input the patient's four diagnostic methods information into the AI ​​Famous Doctor Subsystem, introduce the risk level of inflammation-cancer transformation as a constraint parameter, adjust the diagnosis and treatment analysis process, generate preliminary diagnosis results and medication plans, submit the initial diagnosis plan to the famous doctor inheritance team for proofreading, and generate a plan proofreading log; The process of generating preliminary diagnostic results and medication plans in the risk constraint module is as follows: The patient's four diagnostic methods (inspection, auscultation and olfaction) information is vectorized and encoded by the input processing layer of the AI-powered expert doctor subsystem. This vectorization is then combined with the patient's individual characteristics and historical medical data to generate an input feature vector. The system retrieves the patient's tongue appearance, pulse, symptom description, and lifestyle habits from the database. Textual or categorical information is converted into numerical vectors using techniques such as word segmentation, word embedding, or sparse coding. Simultaneously, individual patient characteristics such as age, gender, and disease course are extracted and standardized. Combined with the patient's historical medical data, including past diagnostic results and medication regimens, historical feature vectors are generated through time-series processing or statistical embedding. Finally, the four diagnostic methods vector, individual feature vector, and historical feature vector are integrated to form a unified input feature vector. This provides a calculable and structured input for the multi-task prediction model, ensuring that the input features are closely related to the patient's actual medical condition and potential risks.

[0028] The input feature vector is fed into the multi-task syndrome differentiation prediction branch and the medication recommendation branch, while the patient's inflammation-cancer conversion risk level is introduced as a constraint parameter. The input feature vector is fed into two neural network branches respectively. The syndrome differentiation prediction branch outputs the category probability of each standard syndrome type to predict the possible syndrome type of the patient. The medication recommendation branch outputs the probability distribution of drug codes and dosages to generate an executable medication plan. At the same time, the patient's inflammation-cancer conversion risk level is encoded as an additional feature and input into the network to adjust the network weights and prediction distribution, so that the model pays more attention to key syndrome differentiation and drug combinations for high-risk patients, and achieves differentiated prediction capability under risk constraints.

[0029] The output results are fused, combining the probabilities of syndrome differentiation and medication recommendations predicted by each branch. Abnormal predictions inconsistent with risk constraints are eliminated, generating preliminary syndrome differentiation results and medication plans. The probabilities of each syndrome type output by the syndrome differentiation branch and the drug probabilities output by the medication branch are weighted and fused, taking into account the constraint effect of risk level on each predicted probability. Predictions that do not conform to risk constraint rules or are abnormal are eliminated or adjusted. Based on the fused probability distribution, the syndrome type with the highest probability is selected as the preliminary syndrome differentiation result. Drug combinations and dosages that meet risk constraints and have high probabilities are selected to generate a preliminary medication plan. The final output of the preliminary syndrome differentiation results and medication plan is recorded in a structured format, including predicted syndrome type codes, drug codes and dosages, prediction confidence levels, and risk level constraint information, providing clinicians with an actionable reference.

[0030] The process of generating the scheme verification log in the risk constraint module is as follows: The preliminary diagnosis and medication plan are submitted to the renowned physician inheritance team. The expert team reviews the preliminary diagnosis category and medication combination in a visual interface and adjusts the plan by selecting, modifying, and adjusting it. The preliminary plan is pushed to the visual interactive interface in a structured format, including the diagnosis category code, drug code and dosage, prediction confidence level, and risk level information. The interface displays the patient's four diagnostic methods information, historical treatment data, and AI prediction results. The expert team can directly view each record in the interface and adjust the diagnosis category and medication combination by selecting, adding, deleting, or modifying to ensure that the plan conforms to clinical experience and actual treatment logic, thereby achieving an effective combination of AI preliminary prediction and human professional experience.

[0031] For each modification operation during the expert review process, the modification rules, logical basis, and associated symptom-syndrome differentiation-medication characteristics are recorded. Each operation by the expert on the syndrome differentiation category or medication combination is automatically captured, including operation type such as addition, deletion, adjustment, the status of the plan before and after the operation, and the operation time. At the same time, each modification is associated with the patient's corresponding symptom characteristics, syndrome differentiation characteristics, and medication characteristic vectors, and the operation logic basis is recorded, such as the reason for the expert to adjust the medication, symptom severity, risk level matching, etc. All operation information is stored in a structured manner to form a complete operation trajectory and decision logic, providing traceable data for knowledge extraction and model optimization.

[0032] The optimized solutions and corresponding expert knowledge enhancement matrices are stored in the disease-specific database, and a solution verification log is generated simultaneously. Each diagnostic result and medication combination after expert verification is stored in the disease-specific database in a structured format, including standardized diagnostic codes, drug codes and dosages, patient's four diagnostic methods information and risk level. At the same time, the knowledge enhancement matrix generated by the expert modification operation records the modification rules, logical basis and correlation characteristics and is stored in association with the optimized solution. A solution verification log is generated simultaneously, recording the preliminary prediction, expert verification operation, optimization results and knowledge enhancement information for each solution, ensuring data integrity and traceability, and providing a foundation for AI model iteration, rule optimization and clinical reference.

[0033] Model update module: Based on the scheme verification log and newly added case data, the syndrome differentiation prediction model and medication recommendation model are iteratively updated to generate the updated AI expert doctor subsystem.

[0034] The process of generating the updated AI expert doctor subsystem in the model update module is as follows: Based on the scheme verification logs, expert modification operations, adjustment logic, and risk constraint application are extracted and transformed into knowledge-enhancing features. Scheme verification logs are obtained from the disease-specific database, and the modification operations of experts on the syndrome differentiation category or medication combination are extracted from each record, including operation type, scheme status before and after addition, deletion, adjustment, and modification, and operation time. The logical basis corresponding to each modification is analyzed, such as symptom correlation, syndrome differentiation rationality, or application of risk level constraints. These modification operations and logical rules are encoded into knowledge-enhancing feature vectors, forming additional inputs that can be used for model training. This ensures that the AI ​​model can learn the application of expert experience and risk constraints during the iteration process, and achieve continuous enhancement of model knowledge.

[0035] The newly added case data is standardized and feature-encoded, and then merged with knowledge-enhanced features to form an iterative training dataset. Information on the four diagnostic methods (palliative diagnoses, auscultation and olfaction, inquiry, and palpation), historical diagnostic results, and medication records of newly added patients are obtained from clinical settings or databases. The information on the four diagnostic methods is structured, converting tongue appearance, pulse appearance, symptom descriptions, and lifestyle habits into a unified coding format. Historical diagnostic results and medication regimens are categorized and standardized, mapped into unified semantic labels. The structured information and individual patient characteristics, such as age, gender, and disease course, are encoded into vectors. Subsequently, these patient feature vectors are merged with knowledge-enhanced feature vectors to construct a complete iterative training dataset. This allows the model to be trained using both newly added case information and expert knowledge, thus maintaining a close correlation between the data and the model's prediction problem.

[0036] Based on the iterative training dataset, a multi-task learning strategy is used to update the parameters of the syndrome differentiation prediction model and the medication recommendation model, while maintaining the risk level constraint adjustment mechanism, generating an updated AI expert doctor subsystem. The iterative training dataset is input into the multi-task neural network, where the syndrome differentiation prediction branch outputs the probability of syndrome type, and the medication recommendation branch outputs the drug code and dosage probability. During training, the prediction results are calculated through forward propagation, and the model parameters are updated through backpropagation. A multi-task loss function is used to balance the training weights of syndrome differentiation prediction and medication recommendation. Risk level constraint features are introduced to adjust the loss weights for high-risk patients during training, ensuring that the model outputs reasonable predictions for key patients. After training, the updated model weights, input processing module, and output interface are encapsulated into a new version of the AI ​​expert doctor subsystem, enabling continuous optimization of the syndrome differentiation prediction and medication recommendation capabilities based on the four diagnostic methods of future patients, while incorporating expert knowledge and risk constraint experience.

[0037] Risk Update Module: Based on the updated AI Expert Doctor Subsystem, the module analyzes patient diagnosis and treatment information in the disease database, updates the corresponding inflammation and cancer transformation risk assessment level, and links and stores the patient diagnosis and treatment data with the updated inflammation and cancer transformation risk level to form a unified patient diagnosis and treatment and risk dataset.

[0038] The process of updating the corresponding inflammation-cancer transformation risk assessment level in the risk update module is as follows: Feature extraction and structured coding are performed on patient information from the four diagnostic methods, historical diagnosis results, and medication regimens in the disease-specific database. Patient tongue appearance, pulse appearance, symptom descriptions, lifestyle habits, historical diagnosis results, and medication regimens are obtained from the disease-specific database. Tongue appearance, pulse appearance, symptoms, and lifestyle habits are standardized and coded, converting textual descriptions into unified category or numerical codes to make them calculable. Historical diagnosis results and medication regimens are categorized and mapped to standard semantic labels, generating vectorized representations. The vectors from the four diagnostic methods, historical diagnosis, and medication are integrated with basic patient characteristics such as age, gender, and disease course to form a unified structured feature vector. This provides executable input for the risk assessment of the AI-powered expert doctor subsystem and ensures that the input features are closely related to the patient's potential risk of cancer transformation.

[0039] The extracted patient features are input into the updated AI expert doctor subsystem. By combining the syndrome differentiation prediction branch and the medication recommendation branch with a risk constraint mechanism, the probability of inflammation-to-cancer transformation for each patient is calculated. The structured feature vector is input into the updated multi-task AI expert doctor subsystem. The syndrome differentiation prediction branch outputs the probability of the patient's syndrome type, and the medication recommendation branch outputs the probability of drug combination and dosage. At the same time, the risk constraint features are embedded in the network to adjust the weighted adjustment of the prediction for high-risk patients, ensuring that the syndrome differentiation and medication results output by the model are associated with the potential risk of inflammation-to-cancer transformation. Based on the network output results, the probability of inflammation-to-cancer transformation for each patient is calculated through a preset risk probability mapping function, realizing the automatic quantitative assessment from structured diagnosis and treatment features to risk probability.

[0040] Based on the calculated cancer conversion probability and the preset risk level mapping rules, the patient's risk level is updated, and a traceable level modification record is generated. The cancer conversion probability of each patient is compared with the preset risk level mapping table, and the probability range is mapped to the corresponding risk level, such as low risk, medium risk, and high risk. The corresponding risk level field of the patient in the database is updated, and the level before the update, the calculated probability value, the mapping rule, and the modification time are recorded. The level modification record is stored in a structured format to form a traceable log for subsequent analysis, model iteration, and clinical decision reference, ensuring that the risk level update process is complete, transparent, and verifiable.

[0041] The process of forming a unified patient diagnosis and treatment and risk dataset in the risk update module is as follows: The system acquires updated patient risk levels for cancer-related inflammation, along with corresponding diagnostic information, syndrome differentiation results, and medication regimens. It extracts the latest diagnostic information for each patient from the disease-specific database, including tongue appearance, pulse, symptom description, and lifestyle habits. It also acquires syndrome differentiation results and medication regimens generated by the AI-powered expert subsystem and verified by experts, and reads the latest updated risk levels for cancer-related inflammation. All data is structured and encoded, converting text descriptions, category information, and risk levels into numerical vectors or standardized codes to ensure that data from different sources and formats are computable and fusionable, providing an executable basis for integrated operations.

[0042] By utilizing the traceable risk level modification records generated during the update process, patient treatment data and historical risk level changes are integrated to form a complete patient record. Historical risk level changes for each patient are extracted from the risk level modification log, including modification time, pre- and post-modification risk levels, modification basis, and related characteristics. This historical risk level information is then matched and integrated with the latest diagnostic information, syndrome differentiation results, and medication regimens to form a multidimensional patient record. Each record contains both current treatment information and the trajectory of risk level changes over time. By encoding the data using time series or identifier fields, it is ensured that historical information and current treatment status clearly correspond within the same structured data object, thereby achieving complete integration of patient treatment and risk data.

[0043] The integrated records are stored in a disease-specific database to form a unified patient diagnosis and risk dataset. Multidimensional structured patient records are stored uniformly by patient ID. Record fields include vectors of four diagnostic methods, codes of syndrome differentiation results, codes of medication regimens, latest risk level, and historical risk change sequences. To ensure data traceability and availability for subsequent analysis, each record is also appended with a timestamp and data source information. Storage and retrieval efficiency are optimized through database indexing or partitioning strategies, making the integrated patient diagnosis and risk dataset a unified basic data source for subsequent AI model iteration, expert-assisted analysis, and clinical decision support.

[0044] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A TCM-based intelligent human-machine fusion diagnostic and treatment system for chronic atrophic gastritis, characterized in that, include: Specialized disease database module: Collects information on the four diagnostic methods, syndrome differentiation results, medication plans and historical cases of famous TCM doctors, and records the corresponding data source and time information. The diagnosis and treatment information is uniformly collected and stored in the chronic atrophic gastritis specialized disease database to form a standardized specialized disease diagnosis and treatment dataset. Association Modeling Module: Based on standardized disease diagnosis and treatment datasets, the module parses diagnosis and treatment texts, constructs the association between symptoms, syndrome differentiation, and medication, and trains syndrome differentiation prediction and medication recommendation algorithms to form an AI expert doctor subsystem that can automatically analyze the four diagnostic methods and output preliminary diagnosis and treatment plans. Risk Constraint Module: Input the patient's four diagnostic methods information into the AI ​​Famous Doctor Subsystem, introduce the risk level of inflammation-cancer transformation as a constraint parameter, adjust the diagnosis and treatment analysis process, generate preliminary diagnosis results and medication plans, submit the initial diagnosis plan to the famous doctor inheritance team for proofreading, and generate a plan proofreading log; Model update module: Based on the scheme verification log and newly added case data, the syndrome differentiation prediction model and medication recommendation model are iteratively updated to generate the updated AI expert doctor subsystem; Risk Update Module: Based on the updated AI Expert Doctor Subsystem, the module analyzes patient diagnosis and treatment information in the disease database, updates the corresponding inflammation and cancer transformation risk assessment level, and links and stores the patient diagnosis and treatment data with the updated inflammation and cancer transformation risk level to form a unified patient diagnosis and treatment and risk dataset.

2. The TCM-based intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis according to claim 1, characterized in that, The process of creating a standardized disease-specific diagnosis and treatment dataset is as follows: The collected information from the four diagnostic methods is structured and converted into a unified coding format, including tongue and pulse information, pulse characteristics, symptom descriptions, and lifestyle habits. The historical diagnostic results and medication plans are categorized and standardized, and the diagnosis and treatment records of different famous doctors are mapped into unified semantic tags; Each record is labeled with its source, timestamp, and basic patient information. The structured information from the four diagnostic methods, the results of syndrome differentiation, the medication plan, and historical cases are uniformly stored in the disease-specific database according to the patient dimension, forming a standardized disease-specific diagnosis and treatment dataset.

3. The TCM-based intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis according to claim 2, characterized in that, The process of establishing the relationship between symptoms, syndrome differentiation, and medication is as follows: Based on a standardized disease diagnosis and treatment dataset, natural language processing is performed on the diagnosis and treatment text to parse symptom descriptions, diagnostic terms, and medication records into computable feature vectors. A multidimensional association matrix is ​​constructed, and symptom characteristics, syndrome differentiation categories and medication labels are encoded accordingly, forming an association weight matrix; Semantic alignment is performed on text data from different sources to unify terminology differences and expression styles, and a standardized symptom-diagnosis-medication mapping relationship is generated.

4. The TCM-based intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis according to claim 3, characterized in that, The process of developing an AI-powered expert doctor subsystem capable of automatically analyzing information from the four diagnostic methods and outputting preliminary treatment plans is as follows: The symptom-syndrome differentiation-medication correlation matrix is ​​vectorized to form structured features that can be input into the model; A multi-task prediction model was constructed, including a syndrome differentiation prediction branch and a medication recommendation branch, while the patient's inflammation-cancer transformation risk level was used as the model input for constraint adjustment; During model training, patient samples of different risk levels are weighted to optimize the model's ability to diagnose and predict medication on high-risk and low-risk samples. The trained model is then packaged into an AI-powered expert doctor subsystem.

5. The TCM-based intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis according to claim 4, characterized in that, The process of generating preliminary diagnosis results and medication plans is as follows: The patient's four diagnostic methods information is vectorized and encoded by the input processing layer of the AI ​​Famous Doctor subsystem, and then combined with the patient's individual characteristics and historical medical data to generate an input feature vector. The input feature vector is fed into the multi-task dialectical prediction branch and the medication recommendation branch, while the patient's inflammation-cancer transformation risk level is introduced as a constraint parameter. The output results are fused, combining the probability of syndrome differentiation and the probability of medication recommendation predicted by each branch, eliminating abnormal predictions that are inconsistent with risk constraints, and generating preliminary syndrome differentiation results and medication plans.

6. The TCM-based intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis according to claim 5, characterized in that, The process of generating the scheme verification log is as follows: The preliminary diagnosis results and medication plan are submitted to the renowned doctors' inheritance team. The expert team reviews the preliminary diagnosis category and medication combination in the visual interface and adjusts the plan by selecting, modifying and adjusting it. For each modification made during the expert review process, record the modification rules, logical basis, and related symptom-diagnosis-medication characteristics; The optimized solution and the corresponding expert knowledge enhancement matrix are stored in the disease-specific database, and a solution verification log is generated synchronously.

7. The TCM-based intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis according to claim 6, characterized in that, The process of generating the updated AI expert doctor subsystem is as follows: Based on the solution verification log, the expert modification operations, adjustment logic, and risk constraint application are extracted and transformed into knowledge-enhanced features. The newly added case data is standardized and characterized, and then merged with knowledge-enhanced features to form an iterative training dataset; Based on the iterative training dataset, a multi-task learning strategy is used to update the parameters of the syndrome differentiation prediction model and the medication recommendation model, while maintaining the risk level constraint adjustment mechanism, to generate the updated AI expert doctor subsystem.

8. The TCM intelligent human-machine fusion diagnosis and treatment system for chronic atrophic gastritis according to claim 7, characterized in that, The process of updating the corresponding inflammation-to-cancer transformation risk assessment level is as follows: Feature extraction and structured coding were performed on patient information from the four diagnostic methods, historical diagnostic results, and medication regimens in the disease-specific database; The extracted patient characteristics are input into the updated AI expert subsystem, and the probability of each patient's inflammation-cancer transformation is calculated by combining the syndrome differentiation prediction branch and the medication recommendation branch with the risk constraint mechanism. Based on the calculated cancer transformation probability and the preset risk level mapping rules, the patient's risk level is updated, and a traceable level modification record is generated.

9. A TCM-based intelligent human-machine fusion diagnostic and treatment system for chronic atrophic gastritis according to claim 8, characterized in that, The process of creating a unified patient diagnosis and risk dataset is as follows: Obtain the updated patient's risk level for cancer transformation, along with corresponding information from the four diagnostic methods, syndrome differentiation results, and medication regimen; By utilizing the traceable risk level modification records generated during the update process, patient treatment data and historical risk level change information are integrated to form a complete patient record; The integrated records are stored in a disease-specific database to form a unified patient diagnosis and treatment and risk dataset.

Images