An ophthalmic traditional Chinese medicine diagnosis and treatment service system and platform

Abstract

The application relates to the technical field of medical artificial intelligence, in particular to an ophthalmology traditional Chinese medicine diagnosis and treatment service system, which comprises a central cloud server and a local deployment module; a plurality of local deployment modules are in communication connection with the central cloud server to form a distributed architecture; the central cloud server comprises a traditional Chinese medicine ophthalmology experience knowledge base and an expert medical case database; the local deployment module comprises a data acquisition module and a diagnosis and treatment module; the data acquisition module is used for acquiring diagnosis and treatment data; the diagnosis and treatment module is used for matching the diagnosis and treatment data with central medical cases in the traditional Chinese medicine ophthalmology expert knowledge base and the expert medical case database, and outputting auxiliary diagnosis data. The application breaks the time and space limitations of expert experience, promotes the extension of high-quality medical resources to the grass-roots level, realizes the intelligentization, standardization and universalization of traditional Chinese medicine ophthalmology diagnosis and medical service, and has important clinical application and popularization value.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence technology, specifically to an ophthalmic traditional Chinese medicine diagnosis and treatment service system and platform. Background Technology

[0002] Ophthalmological diseases are numerous, and many complex and difficult-to-treat conditions (such as complex macular degeneration, refractory glaucoma and optic neuropathy caused by glaucoma, and hereditary retinal diseases) are areas where traditional Chinese medicine (TCM) and TCM-Western medicine have a combined advantage. Using Western medicine alone cannot significantly improve treatment efficacy. However, the diagnosis and treatment of ophthalmological diseases in TCM rely heavily on the clinical experience and diagnostic thinking of senior specialists. This valuable expert experience exists in the form of medical records, which are extremely important knowledge assets in the field of ophthalmology. However, the traditional model of managing and applying expert medical records has many problems, such as:

[0003] Existing outpatient or online consultation systems are mostly general-purpose and lack in-depth integration of specific ophthalmologists' diagnostic and treatment knowledge bases. These systems are unable to intelligently generate targeted consultation paths or preliminary treatment suggestions based on the diagnostic points reflected in the expert's medical records.

[0004] Traditional centralized server architectures exhibit performance bottlenecks, single point of failure risks, and high expansion costs when dealing with high-concurrency access (such as multiple primary hospitals calling services simultaneously), the storage and computation of massive medical case data, and data security and privacy protection between different medical institutions. Summary of the Invention

[0005] This invention aims to enhance the application value of the knowledge base of ophthalmology TCM experts and the medical case database, construct a diagnosis and treatment service architecture, and provide an ophthalmology TCM diagnosis and treatment service system and platform. This will enable the implicit diagnosis and treatment experience of authoritative ophthalmology experts to become explicit and shared with several diagnosis and treatment service institutions, thereby improving the efficiency and level of diagnosis and treatment, and realizing the intelligentization, standardization and universal access to ophthalmology consultation and medical services.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution: Firstly, an ophthalmic TCM diagnosis and treatment service system is provided, including a central cloud server and a locally deployed module; Multiple locally deployed modules communicate with the central cloud server, forming a distributed architecture; The central cloud server includes a knowledge base of traditional Chinese medicine ophthalmology experience and a database of expert medical cases. The local deployment module includes a data acquisition module and a diagnosis and treatment module; The data acquisition module is used to acquire diagnostic and treatment data and to standardize the diagnostic and treatment data to obtain standardized diagnostic and treatment data. The diagnosis and treatment module is used to match the diagnosis and treatment data with the central medical cases in the ophthalmology TCM expert medical case database, and output auxiliary diagnostic data and auxiliary treatment decisions.

[0007] Thanks to the distributed architecture, medical institutions that have deployed local modules can obtain expert-level auxiliary diagnostic data based on patient treatment data acquired in clinical practice, through the TCM ophthalmology experience knowledge base and expert medical case database on the central cloud server. This data can optimize the diagnostic judgments of medical personnel. For patients, they can obtain treatment suggestions from TCM ophthalmology experts at their local medical facility, eliminating the need to visit the TCM ophthalmology experts in their area, saving time and improving the efficiency and effectiveness of treatment.

[0008] Furthermore, the diagnosis and treatment module includes an intelligent agent, which is trained using a large-scale TCM ophthalmology model based on retrieval enhancement. The diagnosis and treatment data and medical cases are input into the intelligent agent, and auxiliary diagnostic data and auxiliary treatment decisions are output.

[0009] By applying intelligent agents to the ophthalmology TCM expert medical system, the intelligent agents are continuously trained through historical expert medical cases. They can output more accurate auxiliary diagnostic data and auxiliary treatment decisions based on the diagnosis and treatment data. Furthermore, as the number of cases increases, the intelligent agents can collect and organize more of the experts' experience, allowing the experts' experience to accumulate and optimize the output results.

[0010] Furthermore, the central cloud server is used to obtain local medical cases from multiple local deployment modules, add the obtained local medical cases as new cases to the expert medical case database, or output the central medical cases in the expert medical case database to the local deployment modules; The local medical cases include diagnostic and treatment data, standardized diagnostic and treatment data, auxiliary diagnostic data, and auxiliary treatment decision-making data.

[0011] Because of its distributed architecture, which includes a central cloud server containing a knowledge base of TCM ophthalmology experience and a database of expert medical cases, this architecture not only provides diagnostic and treatment services but also establishes channels for collecting and sharing medical cases. Local medical cases can be collected from several locally deployed modules through the distributed architecture, enriching the database of TCM ophthalmology expert medical cases. Case collection is more efficient, and the database can receive data from medical service institutions that have joined the architecture, providing a data foundation for improving the level of ophthalmology diagnosis and treatment services. Furthermore, medical service institutions in various regions can also obtain medical cases from the database through locally deployed modules, especially cases not available locally, providing a richer data foundation for future talent training and improving local diagnostic and treatment service levels.

[0012] Furthermore, the local deployment module also includes a local database, which stores local medical cases of the local deployment module; The local database is used to obtain central medical cases from the TCM ophthalmology experience knowledge base and the expert medical case database, or to send the local medical cases to the expert medical case database.

[0013] The local database, in conjunction with the ophthalmology TCM expert medical case database on the central cloud server, enables data sharing and interaction.

[0014] Furthermore, the ophthalmology TCM ophthalmology experience knowledge base and expert medical case database include: symptom label cohort, syndrome type database, treatment method and principle elements, prescription database, TCM ophthalmology commonly used Chinese medicine database, and clinical case database.

[0015] The method for constructing the TCM ophthalmology experience knowledge base and the ophthalmology TCM expert medical case database includes the following steps: Collect medical record data for ophthalmological diagnosis and treatment, including paper medical records and digital data from Western medicine ophthalmological examinations; the digital data from Western medicine ophthalmological examinations includes image data. The text of the paper medical record is identified to obtain the electronic medical record text; The electronic medical record text and Western medicine ophthalmology examination data are extracted in a structured manner, and a vector database and knowledge graph are constructed. The case vectors in the vector database include symptoms, diagnosis results, TCM syndrome types, syndrome differentiation, and TCM medications.

[0016] Secondly, an ophthalmic TCM diagnosis and treatment service platform for medical service institutions is provided, including the ophthalmic TCM diagnosis and treatment service system described in any of the above, and also including a video communication module; The remote consultation module is set up in the local deployment module and is used to establish one-to-one, one-to-many, or many-to-many video communication channels between local deployment modules; it is used for remote diagnosis and online clinical teaching and training.

[0017] When patients receive diagnosis from local medical professionals, they not only receive diagnostic data and auxiliary decision-making suggestions based on the TCM ophthalmology experience knowledge base and expert medical record database, but also receive diagnosis and treatment through one-on-one or one-to-many methods. Local medical professionals can receive training during remote consultations to improve their skills. In addition, through many-to-one and many-to-many video communication channels, other medical professionals who are not in the local area can observe and learn from one expert or multiple online experts, thereby improving the skills of the observing medical professionals. This realizes remote multi-terminal training based on the TCM ophthalmology diagnosis and treatment service system, which can efficiently improve the medical skills of a group of medical professionals.

[0018] Furthermore, it also includes a medication recommendation database, which stores medication recommendations for cases. The medication recommendations are sent to the local deployment module and include the proportions of traditional Chinese medicine, decoction methods, and dosage requirements.

[0019] While providing auxiliary diagnostic data, the ophthalmology TCM diagnosis and treatment service system can also offer medication advice, making medication use more standardized and reaching an expert level.

[0020] Furthermore, it also includes a research platform, which includes a research module set up in a locally deployed module; The research module is used to acquire research data, including diagnostic and treatment data, treatment methods, medication recommendations, and clinical results, and upload the research data to the central cloud server.

[0021] Various medical institutions need to conduct clinical trials when developing traditional Chinese medicine. A research platform, including locally deployed modules, allows for simultaneous clinical trials across multiple institutions. In-house preparations are processed on the same platform; the same medication regimens are used; collected diagnostic and treatment data is standardized and consistent; and treatment methods are maintained under unified expert guidance. Based on this, multiple clinical observations are generated, free from subjective constraints and influences. The obtained clinical trial results are more accurate and do not suffer from differences in efficacy due to variations in pulse patterns or medications for the same disease, differences that are difficult to diagnose. Using a single platform minimizes the influence of subjective factors, rapidly generates research results, and makes it easier to identify the causes of differences, thus improving research efficiency, accuracy, and clinical value.

[0022] Furthermore, the research platform also includes a classification module, which is used to classify the collected research data, and the classification results include the patient's physical condition and the corresponding clinical outcome.

[0023] Because clinical trial data can be collected uniformly through scientific research platforms, it is easier to discover patterns from the collected data, classify it, and find the corresponding drug efficacy for different body types.

[0024] Thirdly, a comprehensive ophthalmic TCM diagnosis and treatment service platform is provided, including an ophthalmic TCM diagnosis and treatment service platform for medical service institutions as described above, and an ophthalmic TCM diagnosis and treatment service system for customers. The customer-facing ophthalmic TCM diagnosis and treatment service system includes a post-discharge management module and an eye health management module. The post-discharge management module is used for follow-up visits to patients after treatment; the eye health management module is used to provide consultation to patients.

[0025] The ophthalmology TCM diagnosis and treatment service platform for medical service institutions primarily focuses on building a system architecture for these institutions. Based on this, it can also provide ophthalmology TCM diagnosis and treatment services to patients. A post-discharge management module allows for follow-up visits with patients after treatment. Medical institutions can use this module to understand patients' symptoms and recovery status after treatment and provide subsequent treatment services. Additionally, patients can use the eye health management module for medical consultations and receive treatment advice, such as obtaining the address or contact information of the nearest relevant medical institution.

[0026] Furthermore, the eye health management module is used to establish a link with ophthalmology clinics and output diagnostic service channels from these clinics based on the patient's symptoms. Because the link between the eye health management module and the ophthalmology clinic is established, patients can directly access the diagnostic service channels of these clinics through the eye health management module and obtain medical services through these channels.

[0027] Furthermore, the post-discharge management module also includes a screening and training system. This system uses VR technology to conduct early eye disease screening and vision rehabilitation training for patients after treatment. Given the characteristics of ophthalmology patients, rehabilitation training is necessary after treatment, and the screening and training system can provide guidance on this training.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows: By constructing a standardized ophthalmology medical record database and a distributed system architecture, this invention breaks the limitation of expert knowledge being confined to a single hospital or a few inheritors. It enables the delivery of expert diagnostic and treatment models to multiple scenarios, such as remote consultations, auxiliary diagnosis in primary hospitals, and medical education, in a high-concurrency, low-latency service format, significantly improving the accessibility of high-quality medical resources. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the structure of an ophthalmology TCM diagnosis and treatment service system in Example 1; Figure 2 This is a typical inheritance architecture as shown in Example 1; Figure 3 This is a structural diagram of a large-scale TCM ophthalmology diagnostic and treatment system based on retrieval enhancement, as shown in Example 1. Figure 4 This is an example of a path for training a model with high-quality ophthalmic data in Example 1; Figure 5 This is a schematic diagram of a typical integrated traditional Chinese and Western medicine diagnosis and treatment service model in Example 1; Figure 6 This is a schematic diagram of the construction of the empowerment platform for medical service institutions in Example 3; Figure 7 This is the architecture of an ophthalmic TCM diagnosis and treatment service platform with an H-shaped structure, as shown in Example 4. Detailed Implementation

[0030] The present invention will now be described in further detail with reference to specific embodiments. However, this should not be construed as limiting the scope of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0031] To address the challenges of sharing expert experience, the lack of in-depth specialist support in general consultation systems, and the performance, security, and cost bottlenecks of centralized servers in traditional ophthalmology, this invention proposes a distributed architecture-based TCM ophthalmology diagnosis and treatment service system and platform. The system focuses on enhancing the application value of a TCM ophthalmology experience knowledge base and expert medical case database, constructing a technical architecture that integrates a central cloud server and multiple locally deployed modules. The central cloud server centrally stores and manages a database containing implicit diagnostic and treatment experience from authoritative experts and medical cases. The locally deployed modules are deployed in various primary healthcare institutions. These modules acquire patient treatment data through a data acquisition module, standardize it, and then intelligently match this standardized data with central medical cases to output auxiliary diagnostic data and treatment decision suggestions. This design enables primary care physicians to receive near-expert-level diagnostic analysis and treatment suggestions in real time during actual consultations, significantly improving the standardization and accuracy of diagnoses. Simultaneously, the distributed architecture effectively addresses the performance challenges posed by high-concurrency access and massive data storage, reduces the risk of single points of failure, and enhances data security and privacy protection through localized processing of critical diagnostic information. Overall, this invention breaks the time and space limitations of expert experience, promotes the extension of high-quality medical resources to the grassroots level, and realizes the intelligentization, standardization, and universal access to ophthalmic TCM consultation and medical services, which has important clinical application and promotion value.

[0032] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0033] Example 1 Firstly, an ophthalmic traditional Chinese medicine diagnosis and treatment service system is provided, the structural diagram of which is shown below. Figure 1 As shown, it includes a central cloud server and a local deployment module; Multiple locally deployed modules communicate with the central cloud server, forming a distributed architecture; The central cloud server includes a knowledge base of traditional Chinese medicine ophthalmology experience and a database of medical cases by traditional Chinese medicine ophthalmology experts. The local deployment module includes a data acquisition module and a diagnosis and treatment module; The data acquisition module is used to acquire diagnostic and treatment data and to standardize the diagnostic and treatment data to obtain standardized diagnostic and treatment data. The diagnostic module is used to match the diagnostic data with the central medical cases in the ophthalmology TCM expert medical case database and output auxiliary diagnostic data.

[0034] Thanks to the distributed architecture, medical institutions that have deployed local modules can obtain expert-level auxiliary diagnostic data based on patient treatment data acquired in clinical practice and a database of medical cases from ophthalmology TCM experts. This allows for the optimization of diagnostic judgments by medical personnel. For patients, they can obtain treatment suggestions from ophthalmology TCM experts locally, eliminating the need to travel to the local clinic, saving time and improving the efficiency and effectiveness of treatment. The distributed architecture also allows for a tiered system for local deployment modules, enabling the establishment of a three-tiered "studio-center-workstation" transmission architecture across multiple medical service institutions nationwide. A typical transmission architecture is as follows: Figure 2 As shown, relying on the established inheritance system, we can quickly enter hospitals, realize a standardized data collection path, build an inheritance network platform, and realize a systematic development platform.

[0035] The Traditional Chinese Medicine Ophthalmology Expert Knowledge Base includes symptom labels, syndrome type databases, prescription databases, and databases of commonly used Chinese medicines for ophthalmology, which, together with the expert medical case database, provide reference for diagnosis and treatment services.

[0036] Furthermore, the diagnosis and treatment module includes an intelligent agent, which is trained using a large-scale TCM ophthalmology model based on retrieval enhancement. The diagnosis and treatment data and medical cases are input into the intelligent agent, and auxiliary diagnostic data is output.

[0037] As a specific embodiment, one type of retrieval-enhanced TCM ophthalmology big data model is a multimodal Transformer diagnostic reasoning model. After acquiring diagnostic data, this model performs a hybrid retrieval in a pre-built TCM ophthalmology experience knowledge base and an expert medical case database to obtain candidate medical case information. This candidate medical case information, along with patient diagnostic data, forms a diagnostic sequence. This diagnostic sequence is then input into the multimodal Transformer diagnostic reasoning model, which outputs TCM syndrome differentiation and prescription. The multimodal Transformer diagnostic reasoning model includes a multimodal Transformer fusion model and a large language model. A structural diagram of a retrieval-enhanced TCM ophthalmology big data model-based syndrome differentiation and treatment system is shown below. Figure 3 As shown.

[0038] The module includes: (1) Text acquisition unit: used to export relevant texts of traditional Chinese medicine ophthalmology from historical paper medical records, books, clinical notes and electronic medical record systems, including chief complaint, present illness history, past medical history, four diagnostic information, diagnosis (traditional Chinese and Western medicine), prescription, follow-up, etc.

[0039] (2) Voice acquisition unit: used to collect the audio recording of consultations in the ophthalmology clinic of traditional Chinese medicine and automatically transcribe it into text through ASR.

[0040] (3) Image and examination data acquisition unit: It interfaces with OCT, fundus camera, perimeter, tonometer and other equipment to acquire structured data and image files.

[0041] (4) Data cleaning unit: performs missing value checks, format standardization, deduplication, error correction and sensitive information desensitization on the above data.

[0042] (5) Structured extraction unit: Extract entities such as symptoms, signs, syndromes, prescriptions, drugs and dosages from the text using the rule and sequence labeling model, and reconstruct them according to the TCM consultation structure such as the "Ten Questions Song".

[0043] Preferredly, the database may include vector database and relational database building modules, vector database and relational database text vectorization and vector database, knowledge graph and relational database.

[0044] As a specific embodiment, the data processing module may include a Graph-RAG retrieval enhancement module, which includes query construction and vector retrieval, knowledge graph subgraph retrieval and R-GCN encoding, multimodal Transformer fusion and CLIP-style alignment.

[0045] 1. Query Construction and Vector Retrieval For new patient input First of all , The query text is concatenated with the relevant four diagnostic methods text. The query vector is obtained through the same embedding function. The candidate set is obtained by retrieving the top K similar cases from the vector database based on the selected index (Flat / HNSW / IVF-PQ). .

[0046] 2. Knowledge Graph Subgraph Retrieval and R-GCN Encoding Centered on the entities involved in the candidate cases, from the knowledge graph Extracting the radius as neighborhood subgraph And it is encoded using a relational graph convolutional network R-GCN: , The overall representation of the subgraph is obtained through multiple updates and graph pooling. .

[0047] 3. Multimodal Transformer fusion and CLIP-style alignment query vector Candidate case text embedding Graph representation and image feature vectors Serialized into a sequence Input a multimodal Transformer and learn cross-modal interaction representations using a self-attention mechanism.

[0048] The retrieval module and the multimodal alignment module can first be based on the retrieval ranking loss. and contrastive learning loss Pre-training is performed, during which the model can be trained independently without updating the parameters of the large model.

[0049] Preferably, it also includes a large language model, based on a general medical large language model, which sequentially employs Prompt-level lightweight tuning, efficient parameter fine-tuning (such as LoRA), and optional full fine-tuning to better adapt the model to the TCM ophthalmology scenario, while performing multi-task joint optimization in syndrome classification and prescription recommendation.

[0050] Preferably, the system also includes an online diagnosis and treatment reasoning module. When a new patient seeks online medical treatment, the system sequentially performs input parsing, vector retrieval, knowledge graph subgraph retrieval, retrieval enhancement and integration, and large model diagnosis and treatment generation steps, outputting TCM syndrome differentiation, explanation of treatment ideas, and prescription recommendations.

[0051] It also includes a doctor verification and feedback learning module. Doctors review and modify the system output, generate the final prescription and treatment record, and the system writes this information back to the knowledge base for continuous learning and correction of the subsequent model.

[0052] The method of the present invention will be described using the auxiliary diagnosis and treatment process for patients with optic nerve atrophy as an example.

[0053] Taking a patient with optic nerve atrophy as an example, the input information is as follows: Western medicine diagnosis is "optic nerve atrophy"; the patient's symptoms include "eyeball atrophy, poor sleep, nocturia, and loose stools"; the tongue appearance is "pale tongue with a thin yellow coating"; and the pulse appearance is "wiry pulse and slow pulse".

[0054] The system execution steps are as follows: (1) Parse the input, form the query text and encode it into a vector. ; (2) Several medical records related to liver and kidney deficiency were quickly retrieved from the vector database using the HNSW index; (3) Extract entities and relational subgraphs such as “optic nerve atrophy - liver and kidney deficiency - kidney and liver tonifying formula - wolfberry / lingzhi / kudzu root / salted dodder seed / acorus tatarinowii” from the knowledge graph, encode them through R-GCN, and present them to the doctor in the form of a graph; (4) The Graph-RAG module fuses the retrieved medical record text with the subgraph representation and uses it as a conditional input to the large language model; (5) The system generates a TCM syndrome differentiation of "Liver and Kidney Deficiency" and provides a recommended prescription of "15g of wolfberry; 15g of Ganoderma lucidum; 30g of kudzu root; 15g of salted dodder seed; 10g of Acorus tatarinowii..." and generates a description of the treatment approach. (6) Doctors make fine adjustments to the dosage and compatibility based on the patient’s specific condition and store the final prescription and feedback in the system for subsequent continuous learning and model correction.

[0055] An example of a path for training a model with high-quality ophthalmic data is as follows: Figure 4 As shown.

[0056] By applying intelligent agents to the ophthalmology TCM expert medical system, the intelligent agents are continuously trained through newly collected medical cases and historical expert medical cases. They can output more accurate auxiliary diagnostic data based on the diagnosis and treatment data. Furthermore, as the number of cases increases, the intelligent agents can collect and organize more of the experts' experience, allowing the experts' experience to accumulate and optimize the output results.

[0057] Furthermore, a central intelligent agent can be set up on the central cloud server to obtain medical records information from various locally deployed modules and train the central intelligent agent. The central intelligent agent obtains a larger training set than the locally deployed modules, which can obtain better model training results. The central intelligent agent and the intelligent agents of the locally deployed modules iterate with each other to improve the overall auxiliary diagnostic capabilities of the system.

[0058] Furthermore, the central cloud server is used to acquire local medical cases from multiple local deployment modules, add the acquired local medical cases as new cases to the ophthalmology TCM expert medical case database, or output the central medical cases from the ophthalmology TCM expert medical case database to the local deployment modules; The local medical cases include the diagnostic and treatment data, standardized diagnostic and treatment data, and auxiliary diagnostic data.

[0059] Because of its distributed architecture, which includes a central cloud server containing a database of medical cases from ophthalmology TCM experts, this architecture not only provides medical services but also establishes channels for collecting and sharing medical cases. Local medical cases can be collected from several locally deployed modules through the distributed architecture, enriching the database and making case collection more efficient. The database can also receive data from medical service institutions that have joined the architecture, providing a data foundation for improving the level of ophthalmology diagnosis and treatment services. Furthermore, medical service institutions in various regions can also obtain medical cases from the database through locally deployed modules, especially cases not available locally, providing a richer data foundation for future talent training and improving local diagnosis and treatment service levels.

[0060] Furthermore, the local deployment module also includes a local database, which stores local medical cases of the local deployment module; The local database is used to retrieve central medical cases from the ophthalmology TCM expert medical case database, or to send the local medical cases to the ophthalmology TCM expert medical case database.

[0061] The local database, in conjunction with the central cloud server's TCM ophthalmology experience knowledge base and expert medical case database, enables data sharing and interaction. Beyond the iterative upgrades of the intelligent agents, the system also features significant database upgrades. By integrating structured and unstructured data, standardized medical case data is constructed, and TCM expert medical cases are organized and stored. New medical cases are continuously added, and new case information is collected. This allows for the discovery of trends and conditions in TCM ophthalmology cases from the collected cases, obtaining macro-level data to facilitate overall intervention and medical recommendations, ultimately improving the eye health of the nation.

[0062] Furthermore, the TCM ophthalmology experience knowledge base and expert medical case database include: symptom label queue, syndrome type database, treatment method and principle elements, prescription database, TCM ophthalmology commonly used Chinese medicine database, and clinical case database.

[0063] The method for constructing the TCM ophthalmology experience knowledge base and the ophthalmology TCM expert medical case database includes the following steps: Collect medical record data for ophthalmological diagnosis and treatment, including paper medical records and digital data from Western medicine ophthalmological examinations; the digital data from Western medicine ophthalmological examinations includes image data. The text of the paper medical record is identified to obtain the electronic medical record text; The electronic medical record text and Western medicine ophthalmology examination data are extracted in a structured manner, and a vector database and knowledge graph are constructed. The case vectors in the vector database include symptoms, diagnosis results, TCM syndrome types, syndrome differentiation, and TCM medications.

[0064] Based on the distributed architecture of this embodiment, medical institutions can not only provide traditional Chinese medicine ophthalmology services, but also realize ophthalmology diagnosis and treatment services combining traditional Chinese and Western medicine. A typical schematic diagram of a diagnosis and treatment service model combining traditional Chinese and Western medicine is shown below. Figure 5 As shown, it includes pre-diagnosis data collection, in-diagnosis treatment plan data, and post-diagnosis health follow-up and re-examination, achieving full coverage of the ophthalmology diagnosis and treatment process.

[0065] Example 2 Example 2 further explains the usage of the large-scale model of traditional Chinese medicine ophthalmology based on retrieval enhancement: The retrieval-enhanced TCM ophthalmology big data model of this invention, following the time sequence from data collection, knowledge construction, model training to online inference and evaluation, can be summarized into the following dynamic steps S1-S8. The steps include: S1, to collect and standardize data on patients in traditional Chinese medicine ophthalmology. S2, which constructs training and testing sets based on standardized cases; S3, construct a vector database and knowledge graph / relational database for TCM ophthalmology diagnosis and treatment; S4, Train the Graph-RAG retrieval model and the multimodal alignment model; S5 performs online dialectical diagnosis and treatment reasoning on a finely tuned large language model to generate TCM syndrome differentiation and prescription recommendations. S6, the doctor verifies the diagnosis and treatment results and generates feedback data; S7 uses the test set to evaluate the system performance in three dimensions: text generation, syndrome differentiation, and prescription recommendation. S8. Deploy the system of this invention in a hospital web environment or remote diagnosis and treatment center to realize a closed loop of intelligent auxiliary diagnosis in traditional Chinese medicine ophthalmology.

[0066] Furthermore, S1: Data Collection and Standardization. Patients fill in basic information and self-described symptoms through the official WeChat account or outpatient pre-consultation system, and upload relevant examination reports; they complete the four diagnostic methods (inspection, auscultation and olfaction, palpation, and olfaction) and ophthalmological equipment examinations at the hospital, and doctors supplement and verify the data in the hospital information system (HIS). All data is aggregated in the data collection and preprocessing module, and after cleaning and standardization, it forms a structured case record.

[0067] S2: Construction of Training and Testing Sets. A labeled dataset was constructed using the medical records of renowned TCM ophthalmologist Liao Pinzheng over the years as the core, combined with cases from other TCM ophthalmology experts. Each sample includes a Western medicine diagnosis. List of patient symptoms Four diagnostic methods information Traditional Chinese Medicine (TCM) syndrome differentiation ,prescription Fields such as time and patient are used to divide the training and test sets.

[0068] S3: Dual Knowledge Base Construction. Based on the aforementioned methods, a vector database and a knowledge graph / relational database are constructed for the training set samples to achieve unified management of multi-source heterogeneous data and to distinguish between explicit knowledge modules (relational database and knowledge graph) and implicit knowledge modules (vector database and model parameters).

[0069] Specifically, each medical record text is represented as a sequence. Embedding functions are defined using a Transformer-based pre-trained language model. , to obtain case vector All case vectors form a set. The data is stored in a vector database that supports ANN retrieval. The similarity function uses cosine similarity. .

[0070] The retrieval index structure in the vector database supports three mainstream modes: Flat exact retrieval, HNSW graph index, and IVF-PQ inverted file plus product quantization, which are used to balance accuracy and efficiency in small-to-medium scale and ultra-large scale scenarios.

[0071] Knowledge Graph and Relational Database: Entities such as "symptoms, disease diagnosis, TCM syndrome, prescription, and single herb" are extracted from each case, and a TCM ophthalmology knowledge graph is constructed based on this. Formally, this knowledge graph is denoted as...

[0072] in: This represents a set of entity nodes (Vertices), including but not limited to patient symptom nodes, disease diagnosis nodes, traditional Chinese medicine syndrome nodes, prescription nodes, and drug nodes; Represents a set of relation types, such as "has symptoms", "belongs to syndrome", "treats diseases", "includes drugs", "drug compatibility", etc. Let the set of triples (Edges) satisfy... Each element Corresponding to a semantic relation "triple", where It is the head entity. This is a tail entity. This establishes the relationship type connecting the two. Simultaneously, the relational database stores patient basic information, quantitative indicators of the four diagnostic methods, ophthalmological examination values, and primary keys and constraints corresponding to the aforementioned entities and relationships in a table structure to support transaction management, version control, and data traceability in the knowledge graph construction.

[0073] S4: Training of the Graph-RAG retrieval enhancement model. A query vector is constructed based on the input "Western medicine diagnosis + patient symptoms + four diagnostic methods data". Candidate medical cases and their associated subgraphs are obtained through vector retrieval and knowledge graph subgraph retrieval. A relational graph neural network and a multimodal Transformer are introduced to complete cross-modal alignment, and a learnable retrieval enhancement module is trained under a unified loss function.

[0074] Specifically, query vector construction and vector retrieval. For new patient input First of all , The query text is concatenated with the relevant four diagnostic methods text. The query vector is obtained through the same embedding function. The candidate set is obtained by retrieving the top K similar cases from the vector database based on the selected index (Flat / HNSW / IVF-PQ). .

[0075] Knowledge Graph Subgraph Retrieval and R-GCN Encoding Centered on the entities involved in the candidate cases, from the knowledge graph Extracting the radius as neighborhood subgraph And it is encoded using a relational graph convolutional network R-GCN: , The overall representation of the subgraph is obtained through multiple updates and graph pooling. .

[0076] Multimodal Transformer fusion and CLIP-style alignment ophthalmic imaging (OCT, fundus photography, etc.) First, perform normalization preprocessing (size uniformity, grayscale / intensity normalization, artifact suppression, etc.), then input it into the image feature extraction network to obtain a vector representation:

[0077] in, For visual encoders such as CNN / ViT, the output is linearly projected and mapped to a dimension consistent with the text / graph representation. This is used for subsequent multimodal fusion to form vectors of the same dimension and incorporate them. .

[0078] query vector Candidate case text embedding Graph representation and image feature vectors Serialized into a sequence The input is a multimodal Transformer, which uses a self-attention mechanism to learn cross-modal interaction representations. The output is then used to obtain a fusion context, which is used to compress and rearrange evidence and serves as conditional input to an LLM.

[0079] To improve the robustness of "symptom / image - syndrome / prescription" matching, this invention draws on the CLIP approach and introduces contrastive learning loss to enhance the symptom / image encoding. Syndrome / Prescription Description Code Alignment within the same semantic space. CLIP-style alignment employs a contrastive learning loss applied during training, acting on symptom / image encoding and syndrome / prescription description encoding to bring them into the same semantic space. Essentially, it's a cross-modal alignment approach based on CLIP. CLIP-style alignment falls under the category of "retrieval enhancement and multimodal alignment model training," and its output is a more robust cross-modal fusion context representation. c This allows LLM to better enhance the context for completing syndrome differentiation and prescription generation.

[0080] As a specific embodiment, the alignment method includes constructing positive sample pairs within a batch. , ,the remaining , For negative samples, a contrastive learning loss in the form of InfoNCE is used. Optimize the multimodal encoder and projection head to align symptoms / images with syndrome / prescription descriptions in the same semantic space.

[0081] The retrieval module and the multimodal alignment module can first be based on the retrieval ranking loss. and contrastive learning loss Pre-training is performed, during which the model can be trained independently without updating the parameters of the large model.

[0082] S5: Multi-level fine-tuning of a large model. Based on a general medical language model, combined with Graph-RAG retrieval to enhance the cross-modal fusion context representation of the model output. The model employs prompt-level lightweight tuning, efficient parameter fine-tuning (such as LoRA), and optional full fine-tuning as needed to better adapt it to TCM ophthalmology scenarios. Simultaneously, it performs multi-task joint optimization in syndrome classification and prescription recommendation. The final diagnosis is generated by the LLM; the Transformer is responsible for outputting the fusion context c, but does not replace the final output of the LLM, supplementing the output of the previous Graph-RAG model as input to the LLM.

[0083] The large-scale model fine-tuning of this invention may include the following three schemes, which can be used individually as alternatives or combined according to specific implementation requirements: 1. Prompt-level lightweight tuning (PROMPT) By designing a structured prompt, the input is split into fixed slots, such as "[Western Medicine Diagnosis]...[Chief Complaint]...[Tongue Appearance]...[Pulse Appearance]...", and constraints are imposed on the output format, enabling the model to generate TCM syndrome differentiation and prescription descriptions under a fixed template. In one implementation, in-context learning is performed using a small number of typical medical case samples without updating the model parameters. In another implementation, a small number of soft cue vectors can be trained and used as learnable prefixes before the model input, requiring only updates to the cue-related parameters without modifying the main weights, thus achieving lightweight adaptation to TCM ophthalmology tasks. 2. Parameter Efficient Fine-Tuning (PEFT) Low-rank adaptation methods such as LoRA are employed to introduce low-rank increments into the attention weight matrix. For the original weights... Define the new weights as ,in , For a trainable low-rank matrix, Only updates during training. , Freezing the original parameters significantly reduces memory usage and the risk of overfitting.

[0084] 3. Full Adjustment In another implementation, a comprehensive fine-tuning of the general model can be performed on a large-scale corpus of TCM ophthalmology medical records to obtain stronger domain fitting capabilities. During comprehensive fine-tuning, the loss function adopts multi-task cross-entropy, which weights and sums the language modeling loss, syndrome classification loss, and prescription multi-label classification loss generated from the medical record description text. For example:

[0085] in, This indicates a loss in the generation of explanatory text for ophthalmology medical records in traditional Chinese medicine. This indicates the loss of classification in TCM syndrome differentiation. This represents the loss in prescription multi-label prediction. , , These are the weighting coefficients for each task, used to balance the importance of different tasks. In one embodiment, full fine-tuning can be performed after efficient parameter fine-tuning has been completed; in another embodiment, full fine-tuning can also be performed directly on the general large model, and this invention does not limit this.

[0086] S6: Online Diagnosis and Treatment Reasoning. When a new patient seeks online treatment, the system sequentially executes the steps of input parsing, vector retrieval, knowledge graph subgraph retrieval, retrieval enhancement and integration, and large model diagnosis and treatment generation, outputting TCM syndrome differentiation, explanation of treatment ideas, and prescription recommendations.

[0087] S7: Doctor Verification and Feedback Learning. Doctors review and modify the system output, generating the final prescription and treatment record. The system writes this information back to the knowledge base and uses it for continuous learning and correction of subsequent models.

[0088] S8: Multi-dimensional evaluation and system deployment. Based on the test set, different retrieval strategies and different basic models are systematically evaluated according to three dimensions: text generation quality, syndrome differentiation effect, and prescription recommendation effect, as well as corresponding multiple indicators. The system of this invention is then deployed in hospitals or remote diagnosis and treatment centers with web environments to realize a closed loop of intelligent auxiliary diagnosis in traditional Chinese medicine ophthalmology.

[0089] In this implementation, retrieval ranking loss and contrastive learning loss This is used for pre-training the retrieval and multimodal alignment modules, while keeping the parameters of the large language model frozen. After training the retrieval and alignment modules, the pre-trained retrieval results are used to construct an enhanced context for multi-task cross-entropy training of the large language model, based solely on... The large language model or its parameters are updated using an efficient fine-tuning layer. Those skilled in the art will understand that the above training process can be implemented on a local computing device or a cloud training platform; this invention is not limited to a specific deployment environment.

[0090] Example 3 The ophthalmology TCM diagnosis and treatment service system in Example 1 primarily addresses the problem of standardizing and degrading the experience of ophthalmology TCM experts through a distributed architecture, thereby enabling the widespread dissemination of expert experience and obtaining expert-level auxiliary diagnostic results. However, based on this system, especially its database and intelligent agent model, it can also be extensively expanded into other applications.

[0091] Example 3 provides an ophthalmology TCM diagnosis and treatment service platform for medical service institutions, including the ophthalmology TCM diagnosis and treatment service system described in any of the above, and also including a video communication module; The remote consultation module is set up in the local deployment module and is used to establish one-to-one, one-to-many, or many-to-many video communication channels between local deployment modules for remote diagnosis and training.

[0092] When patients receive diagnosis from local healthcare professionals, they not only access diagnostic data based on an expert medical record database but also receive diagnosis and treatment through one-on-one or one-to-many methods. Local healthcare professionals can receive training during remote consultations, improving their skills. Furthermore, through many-to-one and many-to-many video communication channels, other healthcare professionals not located locally can observe and learn from a single expert or multiple online experts, thereby improving their own skills. This enables remote multi-terminal training based on the ophthalmology-TCM diagnosis and treatment service system, efficiently enhancing the medical skills of a group of healthcare professionals. Based on a distributed architecture, it provides remote consultation functionality. For complex and difficult cases, experts can directly consult via video, increasing patient access to care and improving service efficiency and quality.

[0093] Furthermore, it also includes a medication recommendation database, which stores medication recommendations for cases. These recommendations are sent to the locally deployed module and include the proportions of traditional Chinese medicine, decoction methods, and dosage requirements. From the perspective of traditional Chinese medicine treatment methods, diagnostic results must be combined with medication to achieve therapeutic effects. Simply providing diagnostic results is insufficient. The ophthalmology traditional Chinese medicine diagnosis and treatment service system provides auxiliary diagnostic data while simultaneously offering medication recommendations, making medication use more standardized and reaching expert-level standards, thus achieving seamless integration between upstream and downstream industries.

[0094] Furthermore, it also includes a research platform, which includes a research module set up in a locally deployed module; The research module is used to acquire research data, including diagnostic and treatment data, treatment methods, medication recommendations, and clinical results, and to upload the research data to the central cloud server.

[0095] The aforementioned improvements to diagnostic pathways and methods are primarily aimed at patients. The platform of this invention can also empower medical service institutions. Various medical service institutions need to conduct clinical trials during the research and development of traditional Chinese medicine. Through a research platform that includes a locally deployed research module, clinical trials can be conducted simultaneously in multiple medical institutions. In-house preparations are on the same platform; the same medication regimen is used; the collected diagnostic and treatment data is standardized and consistent; and treatment methods are consistent through unified expert guidance. Based on this, several clinical observation results are formed, free from the constraints and influences of subjective factors. The obtained clinical trial results are more accurate and do not suffer from differences in efficacy due to different pulse patterns or medications for the same disease, and these differences are difficult to attribute. Using a single platform can minimize the influence of subjective factors, quickly generate research results, and more easily summarize the causes of differences, thus improving research efficiency, accuracy, and clinical value.

[0096] Furthermore, the research platform also includes a classification module, which is used to classify the collected research data, and the classification results include the patient's physical condition and the corresponding clinical outcome.

[0097] Because clinical trial data can be collected uniformly through scientific research platforms, it is easier to discover patterns from the collected data, classify it, and find the corresponding drug efficacy for different body types.

[0098] In summary, based on medical services, the system of this invention can also construct clinical research platforms, talent training platforms, and technology transfer platforms, providing comprehensive B2B services, exploring new productive forces in traditional Chinese medicine, and building a typical specialty construction framework. A schematic diagram illustrating the construction of an empowering platform for medical service institutions is shown below. Figure 6 As shown.

[0099] Example 4 Example 4 provides a comprehensive ophthalmic TCM diagnosis and treatment service platform, including the ophthalmic TCM diagnosis and treatment service platform for medical service institutions described in Example 3, and also includes an ophthalmic TCM diagnosis and treatment service system (C-end) for customers. The customer-facing ophthalmic TCM diagnosis and treatment service system includes a post-discharge management module and an eye health management module. The post-discharge management module is used for follow-up visits to patients after treatment; the eye health management module is used to provide consultation to patients.

[0100] The ophthalmology TCM diagnosis and treatment service platform for medical service institutions primarily focuses on building a system architecture for these institutions. Based on this, it can also provide ophthalmology TCM diagnosis and treatment services to patients. A post-discharge management module allows for follow-up visits with patients after treatment. Medical institutions can use this module to understand patients' symptoms and recovery status after treatment and provide subsequent treatment services. Additionally, patients can use the eye health management module for medical consultations and receive treatment advice, such as obtaining the address or contact information of the nearest relevant medical institution.

[0101] Furthermore, the eye health management module is used to establish a link with ophthalmology clinics and output diagnostic service channels from these clinics based on the patient's symptoms. Because the link between the eye health management module and the ophthalmology clinic is established, patients can directly access the diagnostic service channels of these clinics through the eye health management module and obtain medical services through these channels.

[0102] Furthermore, the post-discharge management module also includes a training system, which uses VR technology to manage the course of treatment and provide vision rehabilitation training for patients after diagnosis and treatment. Given the characteristics of ophthalmology patients, rehabilitation training is necessary after treatment, and the training system can provide guidance on this training.

[0103] The architecture of an H-shaped ophthalmic TCM diagnosis and treatment service platform is as follows: Figure 7 As shown, the foundation (or center) of this H-shaped architecture is a distributed inheritance and application system, and the AI ​​model under this system. For the medical structure end (B end), vertically, it provides a multi-level inheritance architecture of medical service institutions from the province to the city to the district / county to the basic level, promoting the clinical experience of Master Liao Pinzheng of Traditional Chinese Medicine in multiple levels. Horizontally, it builds AI-assisted diagnosis, scientific research platform, talent training platform and achievement transformation platform, providing a platform and foundation for empowering the overall medical service.

[0104] For patients (C-end), it can provide a variety of service models such as Internet hospitals, artificial intelligence systems and eye health service applications. It uses professional expertise to drive non-professional expertise, and provides suitable diagnosis and treatment channels and services for patients with common eye diseases, complex eye diseases, surgical eye diseases and people with eye care needs.

[0105] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A TCM ophthalmic diagnosis and treatment service system, characterized in that, Includes a central cloud server and a locally deployed module; Multiple locally deployed modules communicate with the central cloud server, forming a distributed architecture; The central cloud server includes a knowledge base of traditional Chinese medicine ophthalmology experience and a database of expert medical cases. The local deployment module includes a data acquisition module and a diagnosis and treatment module; The data acquisition module is used to acquire diagnostic and treatment data and to standardize the diagnostic and treatment data to obtain standardized diagnostic and treatment data. The diagnosis and treatment module is used to match the diagnosis and treatment data with the central medical cases in the ophthalmology TCM expert medical case database, and output auxiliary diagnostic data and auxiliary treatment decisions.

2. The ophthalmic TCM diagnosis and treatment service system as described in claim 1, characterized in that, The diagnosis and treatment module includes an intelligent agent, which is trained using a large-scale TCM ophthalmology model based on retrieval enhancement. The diagnosis and treatment data and expert medical cases are input into the intelligent agent, and the agent outputs auxiliary diagnostic data and auxiliary treatment decisions.

3. The ophthalmic TCM diagnosis and treatment service system as described in claim 1, characterized in that, The central cloud server is used to obtain local medical cases from multiple local deployment modules, add the obtained local medical cases as new cases to the expert medical case database, or output the central medical cases in the expert medical case database to the local deployment modules. The local medical cases include diagnostic and treatment data, standardized diagnostic and treatment data, auxiliary diagnostic data, and auxiliary treatment decision-making data.

4. The ophthalmic TCM diagnosis and treatment service system as described in claim 3, characterized in that, The local deployment module also includes a local database, which stores local medical cases of the local deployment module; The local database is used to obtain central medical cases from the TCM ophthalmology experience knowledge base and the expert medical case database, or to send the local medical cases to the expert medical case database.

5. The ophthalmic TCM diagnosis and treatment service system as described in claim 1, characterized in that, The TCM ophthalmology experience knowledge base and ophthalmology TCM expert medical case database include symptom label queue, syndrome type database, treatment method and principle elements, prescription database, TCM ophthalmology commonly used Chinese medicine database, and clinical case database.

6. The ophthalmic TCM diagnosis and treatment service system as described in claim 1, characterized in that, The method for constructing the TCM ophthalmology experience knowledge base and the ophthalmology TCM expert medical case database includes the following steps: Collect medical record data for ophthalmological diagnosis and treatment, including paper medical records and digital data from Western medicine ophthalmological examinations; the digital data from Western medicine ophthalmological examinations includes image data. The text of the paper medical record is identified to obtain the electronic medical record text; The electronic medical record text and Western medicine ophthalmology examination data are extracted in a structured manner, and a vector database and knowledge graph are constructed. The case vectors in the vector database include symptoms, diagnosis results, TCM syndrome types, syndrome differentiation, and TCM medications.

7. An ophthalmic TCM diagnosis and treatment service platform for medical service institutions, characterized in that, The ophthalmic TCM diagnosis and treatment service system as described in any one of claims 1-6 further includes a video communication module; The remote consultation module is set in the local deployment module and is used to establish one-to-one, one-to-many, or many-to-many video communication channels between local deployment modules; Used for remote diagnosis and online clinical teaching and training.

8. The ophthalmology TCM diagnosis and treatment service platform for medical service institutions as described in claim 7, characterized in that, It also includes a medication recommendation database, which stores medication recommendations for cases. The medication recommendations are sent to the local deployment module and include the proportion of Chinese medicine, decoction method, and administration requirements.

9. The ophthalmology TCM diagnosis and treatment service platform for medical service institutions as described in claim 7, characterized in that, It also includes a research platform, which includes a research module set up in a locally deployed module; The research module is used to acquire research data, including diagnostic and treatment data, treatment methods, medication recommendations, and clinical results, and upload the research data to the central cloud server.

10. The ophthalmology TCM diagnosis and treatment service platform for medical service institutions as described in claim 7, characterized in that, The research platform also includes a classification module, which is used to classify the collected research data. The classification results include the patient's physical condition and the corresponding clinical outcome.

11. A comprehensive ophthalmic TCM diagnosis and treatment service platform, characterized in that, The system includes a TCM ophthalmology diagnosis and treatment service platform for medical service institutions as described in any one of claims 6-8, and also includes a TCM ophthalmology diagnosis and treatment service system for customers. The customer-facing TCM ophthalmology diagnosis and treatment service system includes a post-discharge management module and an eye health management module. The post-discharge management module is used for follow-up visits to patients after treatment, and the eye health management module is used to provide consultation to patients.

12. The comprehensive ophthalmic TCM diagnosis and treatment service platform as described in claim 11, characterized in that, The eye health management module is used to establish a link with ophthalmology diagnosis and treatment service institutions and output the diagnostic service channels of the diagnosis and treatment service institutions based on the patient's symptoms.

13. The comprehensive ophthalmic TCM diagnosis and treatment service platform as described in claim 11, characterized in that, The post-discharge management module also includes a screening and training system, which uses VR to conduct early eye disease screening and vision rehabilitation training for patients after treatment.

Images