Ophthalmologic nursing scheme generation method and system based on multiple agents, electronic equipment and storage medium

Abstract

The invention provides a multi-agent-based ophthalmic nursing scheme generation method and system, electronic equipment and a storage medium, and relates to the technical field of medical care application. According to the invention, a plurality of special AI agents coordinated by an agent center are integrated, and all-round ophthalmic nursing from initial symptom capture, image interpretation to evidence retrieval, treatment planning and recording and storage is covered. The system mainly comprises a dialogue agent used for patient interaction, a visual agent used for image analysis, an RAG agent used for accessing latest medical literature, and a guardrail agent used for ensuring safety and interpretability. The technology comprises dual input and output protection measures, mandatory manual verification and role-based access control for patients, doctors and administrators, has expandability, can be deployed locally or in the cloud by utilizing interchangeable components and micro-servers, provides end-to-end workflows for ophthalmology nursing, and improves the working efficiency of ophthalmology nursing. Diagnosis accuracy is high, and treatment effect and patient management effect are good.

Description

Technical Field

[0001] The present invention relates to the field of medical care application technology, and in particular to a multi-agent-based ophthalmic care program generation method, system, electronic equipment and storage medium. Background Art

[0002] Eye diseases and disorders affect millions of people worldwide, compromising quality of life and potentially leading to vision loss if left untreated. Over the years, the field of ophthalmology has made significant advances in diagnostic techniques and treatment modalities. However, challenges remain in providing comprehensive, accurate, and timely eye care, particularly in areas with limited access to specialized ophthalmologists.

[0003] Traditional ophthalmic care workflows typically involve multiple steps, including initial symptom assessment, image interpretation, diagnosis, treatment planning, and follow-up care, which are time-consuming and may lead to delays in diagnosis and treatment initiation. In addition, the increasing amount of medical data, including high-resolution retinal images and complex patient histories, poses challenges in efficient analysis and interpretation. Although artificial intelligence (AI) has become a promising tool in healthcare, it has potential applications in various medical specialties, including ophthalmology. However, existing AI solutions in ophthalmology generally focus on specific tasks, such as image classification or disease detection, and do not address the full range of ophthalmic care needs. In addition, telemedicine and remote monitoring have also become increasingly prominent in recent years, providing potential solutions for improving the accessibility of ophthalmic care services, especially in underserved areas. However, challenges remain in ensuring the accuracy and reliability of remote assessments and seamlessly integrating these technologies into existing healthcare systems.

[0004] As can be seen, current ophthalmic disease diagnosis and management systems often rely on isolated AI models or standalone applications, failing to cover the full range of ophthalmic care. These fragmented solutions lead to inefficiencies, potential errors, and an inconsistent patient experience. Furthermore, existing technologies can struggle with transparency, explainability, and staying current with the latest medical literature.

[0005] Based on this, this application aims to create a comprehensive, integrated solution that addresses these limitations with the goal of improving security, scalability, and user-specific functionality throughout the ophthalmic care workflow. Summary of the Invention

[0006] (1) Technical problems solved

[0007] In response to the shortcomings of the existing technology, the present invention provides a multi-agent-based ophthalmic care plan generation method, system, electronic device and storage medium, which at least solves one or more of the above-mentioned problems, including poor safety, low efficiency and poor patient experience.

[0008] (2) Technical solution

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0010] In a first aspect, the present invention first proposes a multi-agent-based ophthalmic care plan generation method, the method comprising:

[0011] Coordinate multi-agent interactions to process pre-acquired patient data; the multi-agents include at least a conversational agent for patient interaction, a visual agent for image analysis, and a retrieval-enhanced generative agent for accessing the latest medical literature;

[0012] Verifying the output results of a single agent in the multi-agent;

[0013] Review and annotate the verified output results;

[0014] Generate final eye care plan based on the reviewed and approved output results.

[0015] In one embodiment, the patient data includes at least symptom description data, medical image data, and electronic health record data.

[0016] In a preferred embodiment, the method further comprises:

[0017] Processing the symptom description data using an audio processor; processing the electronic health record data using a text processor; and processing the medical image data using an image processor;

[0018] The symptom description data, electronic health record data, medical image data, and metadata related to the symptom description data, electronic health record data, and medical image data are embedded and stored in a vector database.

[0019] In one embodiment, the method further comprises: utilizing a web search agent to retrieve the latest medical literature online.

[0020] In one embodiment, the method further comprises: implementing double input and output protection measures when verifying the output result of a single agent in the multi-agent.

[0021] In one embodiment, when reviewing and annotating the verified output result, the clinician is allowed to edit and output the verified output result through human-computer interaction before the verified output result is output.

[0022] In one embodiment, the method further comprises securely transmitting the final eye care plan to an appropriate recipient using encryption protocols and user authentication measures.

[0023] In a preferred embodiment, a network voice UI is utilized to deliver diagnosis and treatment recommendations through voice interaction.

[0024] In a preferred embodiment, the format and content of the final recommendation output are customized according to the intended recipient.

[0025] In a preferred embodiment, the final recommendations are securely transmitted to the appropriate recipients using encryption protocols and user authentication measures.

[0026] In a preferred embodiment, the identity of a user attempting to access EyeGI is verified using a username and password combination, biometric data, or multi-factor authentication technology.

[0027] In a second aspect, the present invention further proposes a multi-agent-based ophthalmic care plan generation system, the system comprising:

[0028] A data processing module configured to coordinate multi-agent interactions to process pre-acquired patient data; the multi-agents include at least a conversational agent for patient interaction, a visual agent for image analysis, and a retrieval-enhanced generative agent for accessing the latest medical literature;

[0029] an output result verification module, configured to verify the output result of a single agent in the multi-agent;

[0030] The output result approval module is configured to review and annotate the verified output results;

[0031] The final output result generating module is configured to generate a final ophthalmic care plan based on the output results after review and approval.

[0032] In a third aspect, the present invention also proposes an electronic device comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, the steps of the multi-agent-based ophthalmic care plan generation method described above are implemented.

[0033] In a fourth aspect, the present invention finally proposes a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-agent-based ophthalmic care plan generation method as described in any one of the above items.

[0034] (3) Beneficial effects

[0035] The present invention provides a multi-agent-based ophthalmic care plan generation method, system, electronic device, and storage medium. Compared with the existing technology, it has the following advantages:

[0036] This paper provides a multi-agent-based ophthalmic care plan generation method. This method coordinates the interaction of multiple agents, including a dialogue agent, a visual agent, and a retrieval and enhancement generation agent, to process patient data. The method then verifies the output of individual agents within the multi-agent system, reviews and annotates the verified output, and finally generates a final ophthalmic care plan based on the approved output. This method simplifies and enhances the entire ophthalmic disease management process, providing greater safety, efficiency, and a better patient experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 Flowchart of a method for generating an ophthalmic care plan based on multi-agents in an embodiment of the present invention;

[0039] Figure 2 Flowchart of a method for processing medical data using artificial intelligence in an embodiment of the present invention;

[0040] Figure 3 A flowchart of processing patient input in a medical system according to an embodiment of the present invention;

[0041] Figure 4 A flowchart of processing verified artificial intelligence output in a medical system in an embodiment of the present invention;

[0042] Figure 5 Flowchart of the user access control process in the system according to an embodiment of the present invention;

[0043] Figure 6 A block diagram of a multi-agent-based ophthalmic care plan generation system according to an embodiment of the present invention;

[0044] Figure 7 A flowchart of a data flow process for processing medical information in an embodiment of the present invention;

[0045] Figure 8 is a block diagram of an EyeGI system with multiple processing layers according to an embodiment of the present invention;

[0046] Figure 9 is a block diagram of a multi-agent artificial intelligence framework for ophthalmic disease management in an embodiment of the present invention;

[0047] Figure 10is a block diagram of a user access control system according to an embodiment of the present invention;

[0048] Figure 11 1 is a block diagram of a user role-based access control system for a medical platform according to an embodiment of the present invention;

[0049] Figure 12 is a flow chart of the interaction between components in the medical data processing system according to an embodiment of the present invention;

[0050] Figure 13 is a sequence diagram of the workflow of the medical diagnosis system in an embodiment of the present invention;

[0051] Figure 14 is a flow chart of a medical data processing sequence in an embodiment of the present invention;

[0052] Figure 15 It is a sequence diagram of interactions in the medical diagnosis system in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0054] Eye diseases and disorders affect millions of people worldwide, compromising quality of life and potentially leading to vision loss if left untreated. Over the years, the field of ophthalmology has made significant advances in diagnostic techniques and treatment modalities. However, challenges remain in providing comprehensive, accurate, and timely eye care, particularly in areas with limited access to specialized ophthalmologists.

[0055] Traditional eye care workflows typically involve multiple steps, including initial symptom assessment, image interpretation, diagnosis, treatment planning, and follow-up care. This can be time-consuming and can lead to delays in diagnosis and treatment initiation. Furthermore, the ever-increasing volume of medical data, including high-resolution retinal images and complex patient histories, poses challenges in efficient analysis and interpretation.

[0056] With the development of artificial intelligence (AI), AI has become a promising tool in healthcare, with potential applications in various medical specialties, including ophthalmology. AI technologies such as machine learning and computer vision have demonstrated the ability to analyze medical images, recognize patterns, and assist in clinical decision-making. However, integrating AI into comprehensive eye care systems that cover the entire patient journey remains an active area of research and development. Currently, existing AI solutions in ophthalmology typically focus on specific tasks, such as image classification or disease detection, without addressing the full range of eye care needs. There is growing interest in developing more comprehensive AI frameworks that can assist healthcare providers throughout the entire patient care pathway, from initial triage to long-term management.

[0057] Furthermore, telemedicine and remote monitoring have gained prominence in recent years, offering potential solutions for improving access to eye care services, particularly in underserved areas. However, challenges remain in ensuring the accuracy and reliability of remote assessments and seamlessly integrating these technologies into existing healthcare systems.

[0058] As the field of artificial intelligence in healthcare continues to evolve, the demand for innovative approaches that can improve the efficiency, accuracy, and accessibility of eye care services continues to exist. The development of comprehensive AI frameworks that can support the clinical practice of ophthalmologists and other eye care professionals while improving patient outcomes remains an area of active research and development in the field.

[0059] In summary, current ophthalmic disease diagnosis and management systems often rely on isolated AI models or standalone applications, failing to cover the full range of ophthalmic care. These fragmented solutions lead to inefficiencies, potential errors, and a fragmented patient experience. Furthermore, existing technologies can struggle with transparency, explainability, and staying current with the latest medical literature.

[0060] Based on this, this application aims to create a comprehensive, integrated solution that addresses these limitations with the goal of improving security, scalability, and user-specific functionality throughout the ophthalmic care workflow.

[0061] The embodiments of the present application provide a multi-agent-based ophthalmic care plan generation method, system, electronic device and storage medium, which at least solves one or more of the above-mentioned problems, including poor security, low efficiency and poor patient experience, and achieves the purpose of providing an end-to-end workflow for ophthalmic care work, improving diagnostic accuracy, treatment effects and good patient management.

[0062] The technical solution in the embodiments of the present application is to solve the above technical problems, and the overall idea is as follows:

[0063] In order to solve the above-mentioned problems, this application proposes an EyeGI system, a multi-agent artificial intelligence framework for comprehensive ophthalmic disease management. The system integrates multiple specialized artificial intelligence agents coordinated by a central hub, covering a full range of ophthalmic care from initial symptom capture and image interpretation to evidence retrieval, treatment planning and record keeping. The main components include: a conversational agent for patient interaction, a visual agent for image analysis, a RAG (retrieval-augmented generation) agent for accessing the latest medical literature; and a guardrail agent for ensuring security and explainability. The EyeGI system includes dual input and output protection measures, mandatory manual verification, and role-based access control for patients, doctors, and administrators. At the same time, EyeGI is designed with scalability in mind. It uses interchangeable components and microservers and can be deployed locally or in the cloud to provide an end-to-end workflow for ophthalmic care, improving diagnostic accuracy, treatment effectiveness, and overall patient management.

[0064] It is important to note that EyeGI, proposed in this application, can be implemented in ophthalmology clinics, hospitals, and eye care centers to streamline and enhance the entire ophthalmic disease management process. By integrating this multi-agent AI framework into existing healthcare systems, healthcare institutions can significantly improve their diagnostic accuracy, treatment planning, and overall patient care efficiency. More specifically, one goal of this application is to propose an integrated eye care platform that eliminates the need for multiple disconnected systems, thereby reducing operating costs and improving workflow efficiency across the entire eye care process. The EyeGI framework can be used to automate initial patient triage, assist with image interpretation, provide evidence-based treatment recommendations, and seamlessly integrate with electronic medical records, while maintaining high standards of security and interpretability. Furthermore, EyeGI's scalability and vendor-neutral nature allow its implementation in a variety of healthcare settings, from small private clinics to large hospital networks. This versatility opens up significant commercial opportunities in the medical technology market, potentially revolutionizing the way eye care is delivered and managed globally.

[0065] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0066] Example 1:

[0067] In the first aspect, the present invention first proposes a method for generating an ophthalmic care plan based on multi-agents, see Figure 1 , the method comprising:

[0068] S1. Coordinate multi-agent interactions to process pre-acquired patient data; the multi-agents include at least a conversational agent for patient interaction, a visual agent for image analysis, and a retrieval-enhanced generative agent for accessing the latest medical literature;

[0069] S2. Verify the output result of a single agent in the multi-agent;

[0070] S3. Review and annotate the verified output results;

[0071] S4. Generate the final ophthalmic care plan based on the reviewed and approved output results.

[0072] The following combination Figure 1-5 , and explanations of the specific steps of steps S1-S4 are provided to describe in detail the implementation process of an embodiment of the present invention.

[0073] S1. Coordinate multi-agent interaction to process pre-acquired patient data; the multi-agent includes at least a conversational agent for patient interaction, a visual agent for image analysis, and a retrieval-enhanced generation agent for accessing the latest medical literature.

[0074] When processing patient data, a multi-agent center is set up as a central processing unit to coordinate and manage multiple agents dedicated to analyzing and processing different aspects of ophthalmology-related medical data, such as a dialogue agent, a visual agent, and a retrieval-augmented generation (RAG) agent to process patient data.

[0075] An agent (artificial intelligence agent) is a system based on artificial intelligence technology that can perceive its environment, reason, make decisions, and take actions to achieve specific goals. In this embodiment, a conversational agent is used for patient interaction, responsible for processing and analyzing text-based patient data such as symptom descriptions or medical histories. The visual agent focuses on analyzing medical images related to eye diseases, and the RAG agent is responsible for retrieving and analyzing relevant medical information from a knowledge base.

[0076] In a preferred embodiment, the multi-agent system also includes a web search agent that is used to retrieve the latest medical literature from online sources. The web search agent supplements the RAG agent by providing access to the latest research, clinical trials, or treatment guidelines that have not yet been included in the system's internal knowledge base.

[0077] Before processing patient data, it is necessary to pre-acquire the patient data, which includes at least symptom descriptions, medical images, and multiple types of data from electronic health records. Preferably, when acquiring the above patient data, an audio processor processes the patient's verbal description of symptoms or medical history; a text processor processes written information from electronic health records or patient questionnaires; and an image processor manages and pre-processes medical images, such as retinal scans or OCT images.

[0078] After acquiring the patient data, the patient data is stored as embeddings in a vector database, such as Figure 2 Preferably, when storing the patient data as embedding, the method further includes storing metadata related to the patient data embedding in the vector database.

[0079] S2. Verify the output results of a single agent in the multi-agent.

[0080] To ensure that the insights generated by each agent after processing patient data are reliable and compliant with established medical guidelines, they need to be validated before patient care decisions are made.

[0081] In one embodiment, Figure 2 As shown, the Guardrails agent applies predefined rules and constraints to the output of the multi-agent hub, which are based on established medical guidelines, ophthalmology best practices, or regulatory requirements. The Guardrails agent flags any recommendations that exceed these predefined parameters for further review or correction. Additionally, the Guardrails agent performs consistency checks on the different components of the AI-generated output. Furthermore, the Guardrails agent incorporates safety thresholds for certain types of recommendations. In some cases, if an AI-generated recommendation exceeds a predefined safety threshold, the Guardrails agent automatically flags it for mandatory human review or requires the AI system to provide additional justification before allowing it to proceed.

[0082] Furthermore, when verifying the output of specialized agents, dual input and output safeguards are implemented within the guardrail agent. "Dual input and output" generally refers to the presence of two different types or two independent input channels and two corresponding output channels within a system, device, or process. Dual input and output safeguards can improve reliability, increase flexibility, and enhance processing capabilities.

[0083] S3. Review and annotate the verified output results.

[0084] The output results after verification in step S2 need to be further reviewed, modified and approved to ensure that the verified output results are, firstly, technically reasonable and in line with medical guidelines, and secondly, consistent with the judgment of experienced medical professionals.

[0085] In one embodiment, when reviewing and annotating the verified output results, clinicians are allowed to edit the verified output results through human-computer interaction before the verified output results are output. For example, a platform is provided for medical professionals through a manual review interface to review, modify and approve the AI-generated recommendations before they are finalized and communicated to patients or other medical service providers. Figure 2 、 Figure 3 、 Figure 4 shown.

[0086] Preferably, the human review interface presents the AI-generated recommendations in a structured and easy-to-understand format; and / or the human review interface enables medical professionals to interact with the AI-generated output and allows clinicians to edit the output before approving it; and / or the human review interface also provides access to the underlying data and analysis used by the AI system to generate its recommendations; and / or the human review interface includes a feedback mechanism that allows clinicians to provide feedback on the quality and accuracy of the AI-generated recommendations.

[0087] S4. Generate the final ophthalmic care plan based on the reviewed and approved output results.

[0088] Once the AI-generated outputs are verified by the guardrail agent in step S2 and approved by the human review interface in step S3, they will be communicated to the appropriate recipients, such as patients or other healthcare providers, to generate final diagnosis and treatment recommendations based on the approved outputs. Figure 4 shown.

[0089] In some preferred embodiments, the output of the final recommendations is formatted in a standardized structure to ensure consistency and ease of interpretation, and is based on a FHIR REST API, allowing the output module to seamlessly integrate with other medical information systems and electronic health records. Preferably, the Open EMR EDC interface is utilized to facilitate direct integration of EyeGI output with electronic medical record systems.

[0090] In some embodiments, a network voice UI is utilized to deliver diagnosis and treatment recommendations through voice interaction.

[0091] In some embodiments, the format and content of the final recommendation output may also be customized based on the intended recipient.

[0092] In some embodiments, the final recommendations are securely transmitted to the appropriate recipients using encryption protocols and user authentication measures.

[0093] In some preferred embodiments, various methods are used to verify the identity of users attempting to access EyeGI, such as username and password combinations, biometric data, or multi-factor authentication technology. More preferably, roles such as patient, doctor, or administrator are assigned based on predefined criteria or information stored in the system. In addition, specialized function modules are set up for different user types to control access to specific functions related to the corresponding user role. Figure 5 shown.

[0094] In other preferred embodiments, EyeGI activity and access events are logged, creating an audit trail of user interactions with EyeGI. This logging functionality is useful for security monitoring, compliance purposes, and system performance analysis. Figure 5 shown.

[0095] In further embodiments, administrators are allowed to modify system settings, update user roles, and manage access rights. Figure 5 shown.

[0096] Example 2:

[0097] This embodiment proposes a multi-agent-based ophthalmic care plan generation system, which includes a multi-agent artificial intelligence framework (referred to as EyeGI) for ophthalmic disease diagnosis and management. This multi-agent artificial intelligence framework integrates multiple specialized artificial intelligence agents that execute the corresponding steps of the method in Example 1 above, covering the entire ophthalmic care process, such as initial symptom capture, image interpretation, evidence retrieval, treatment planning, and record keeping.

[0098] The EyeGI proposed in this embodiment includes various agents, including a conversational agent for interacting with patients, a visual agent for image analysis, a retrieval-augmented generation (RAG) agent for accessing the latest medical literature, and a guardrail agent for ensuring safety and explainability. This multi-agent AI framework is designed for scalability, utilizing interchangeable components and microservers, and can be deployed locally or in the cloud.

[0099] In some cases, EyeGI provides an end-to-end workflow for ophthalmic care, improving diagnostic accuracy, treatment outcomes, and overall patient management. EyeGI employs dual input and output safeguards, as well as mandatory human verification, to ensure the reliability and security of its output.

[0100] In some cases, EyeGI implements role-based access control for patients, physicians, and administrators, allowing appropriate levels of system interaction and data access based on user roles. This approach may help maintain data privacy and system security while enabling efficient collaboration between different stakeholders in the eye care process.

[0101] By leveraging artificial intelligence and a multi-agent architecture, EyeGI offers potential benefits such as improved diagnostic accuracy, streamlined workflows, and enhanced decision support for eye care professionals. EyeGI also integrates the latest medical knowledge into the diagnosis and treatment planning process through its real-time information retrieval capabilities.

[0102] EyeGI provides a comprehensive framework for the diagnosis and management of eye diseases. Figure 6 As shown, Figure 6 This section describes the EyeGI system block diagram, which may include multiple interconnected components arranged in a sequential flow structure. Specifically:

[0103] EyeGI begins with an input module connected to a vector database. The input module receives various types of patient data, such as symptoms, medical images, and electronic health records. The vector database may store and manage the processed input data in a format suitable for efficient retrieval and analysis.

[0104] The Agent Center forms the central processing unit of EyeGI. It contains multiple specialized agents, each designed to perform a specific task within the eye care workflow. These agents work together to process and analyze data, providing comprehensive insights for diagnosis and treatment planning.

[0105] Following the Agent Center, EyeGI includes a Guardrail Agent. The Guardrail Agent acts as a safety mechanism that verifies the outputs from the Agent Center to ensure reliability and compliance with established medical guidelines.

[0106] A human review interface has been incorporated into EyeGI to allow for expert oversight and intervention. This interface enables clinicians to review and approve the system's recommendations, ensuring that healthcare professionals' expertise remains a critical part of the diagnostic and treatment process.

[0107] EyeGI ends with an output module at the end of the processing chain. The output module generates final recommendations, treatment plans and other relevant information based on the processed and validated data.

[0108] In some embodiments, EyeGI utilizes microservers that can run locally or in the cloud. This architecture provides flexibility in deployment options, allowing healthcare providers to choose the most appropriate infrastructure based on their specific needs and resources. The use of microservers also enhances the scalability and maintainability of the system, making it easier to update and modify individual components without affecting the entire system.

[0109] EyeGI includes input modules and data processing components to handle various types of patient data. Figure 7 As shown, Figure 7 The data flow process for processing medical information is described, including multiple stages from patient data input to output generation.

[0110] In one embodiment, the input module may receive different types of patient data, such as symptom descriptions, medical images, and electronic health records. The input module is part of the data ingestion layer, such as Figure 8 As shown, the data capture layer includes specialized processors for processing different data types.

[0111] The data capture layer includes audio processors, text processors, and image processors. The audio processor processes verbal descriptions of symptoms or medical history provided by the patient. The text processor processes written information from electronic health records or patient questionnaires. The image processor manages and preprocesses medical images, such as retinal scans or OCT images. Figure 9 shown.

[0112] After initial processing at the data capture layer, the processed data is stored in a vector database. This database stores vector embeddings of the processed data, enabling efficient retrieval and analysis by the agent center. In some cases, the database also stores metadata in addition to the vector embeddings. This metadata includes information such as the data source, timestamp, patient identifier, or other relevant contextual information that can be useful for subsequent analysis or retrieval.

[0113] The vector database serves as a central repository for all processed patient data, enabling rapid access and efficient querying across EyeGI's components. Using vector embeddings allows for semantic similarity searches and potentially facilitates the integration of different data types within a unified framework.

[0114] In some cases, the vector database interfaces bidirectionally with the agent center, e.g. Figure 8 This bidirectional connection allows specialized agents within the agent center to retrieve relevant data for analysis and also enables the agents to update or enrich the stored data based on their processing results.

[0115] Data flow process and Figure 8 The architecture described in [1] demonstrates how EyeGI handles the ingestion, processing, and storage of diverse patient data types. This approach has the potential to provide the foundation for comprehensive eye disease diagnosis and management, ensuring that all relevant patient information is appropriately processed and available for analysis by specialized AI agents within the system.

[0116] EyeGI consists of a multi-agent center, such as Figure 8 As shown in Figure 2, the multi-agent center acts as a central processing unit, coordinating and managing multiple agents specifically designed to analyze and process ophthalmology-related medical data.

[0117] In one embodiment, the multi-agent center includes a conversation agent, a vision agent, a RAG agent, and a web search agent. Each of these agents performs a specific function within EyeGI, contributing to comprehensive ophthalmic disease diagnosis and management.

[0118] Conversational agents process and analyze text-based patient data, such as symptom descriptions or medical histories. In some cases, they interact directly with patients, asking relevant questions and interpreting responses to gather the necessary information for diagnosis.

[0119] Vision agents focus on analyzing medical images related to eye diseases. In some cases, they process retinal scans, OCT images, or other ophthalmic imaging data. Along with their analysis results, the agents return heatmaps, providing visual representations of areas of interest or focus within the image. These heatmaps enhance the interpretability of the agent's findings, enabling medical professionals to better understand and interpret the AI-generated results.

[0120] The RAG agent is responsible for retrieving and analyzing relevant medical information from a knowledge base. In some cases, the RAG agent provides inline citations when presenting information, enhancing the interpretability and traceability of its output. This functionality allows medical professionals to verify the source of information used in the diagnostic process and stay informed about the latest research and guidelines in the field of ophthalmology.

[0121] The web search agent is designed to retrieve the latest medical literature from online sources. In some cases, the web search agent complements the RAG agent by providing access to the latest research, clinical trials, or treatment guidelines that have not yet been incorporated into the system's internal knowledge base. This capability helps ensure that EyeGI stays current with the latest developments in eye care and ophthalmology.

[0122] The multi-agent center coordinates the activities of these specialized agents, managing the flow of information between them and integrating their outputs. The multi-agent center receives processed data from the vector database and distributes relevant information to each agent for analysis. The multi-agent center then combines and synthesizes the results from each agent to generate comprehensive diagnostic insights and treatment recommendations.

[0123] The output of the multi-agent center is sent to the guardrail agent for validation before being presented through a human review interface. This process ensures that the AI-generated insights are reliable, align with established medical guidelines, and receive expert oversight before being used to make patient care decisions.

[0124] EyeGI includes guardrail agents and human review interfaces, such as Figure 8 These components play a critical role in ensuring the safety, accuracy, and reliability of the system output.

[0125] In some embodiments, the guardrail agent acts as a validation mechanism for the outputs generated by the multi-agent hub. The guardrail agent is designed to perform various checks and validations on AI-generated recommendations before they are presented to human review.

[0126] In some embodiments, the guardrail agent applies predefined rules and constraints to the output of the multi-agent hub. These rules are based on established medical guidelines, ophthalmology best practices, or regulatory requirements. The guardrail agent flags any recommendations that fall outside these predefined parameters for further review or correction.

[0127] In some embodiments, the Guardrail Agent performs consistency checks on the different components of the AI-generated output. For example, the Guardrail Agent compares the Vision Agent's recommendations with the RAG Agent's recommendations to ensure there are no contradictions or inconsistencies in the proposed diagnosis or treatment plan.

[0128] In some embodiments, the Guardrails agent also incorporates safety thresholds for certain types of suggestions. In some cases, if an AI-generated suggestion exceeds a predefined safety threshold, the Guardrails agent automatically flags it for mandatory human review or may require the AI system to provide additional justification before allowing it to proceed.

[0129] After validation by the Guardrail Agent, the output is sent to a human review interface, which provides a platform for medical professionals to review, modify, and approve the AI-generated recommendations before they are finalized and communicated to patients or other healthcare providers.

[0130] In some embodiments, a human review interface presents AI-generated recommendations in a structured and easy-to-understand format. The interface may display key findings, recommended diagnoses and treatment plans, and supporting evidence and explanations provided by individual agents in the multi-agent hub.

[0131] In some embodiments, a human review interface allows medical professionals to interact with AI-generated outputs. In some cases, clinicians use the interface to request additional information, modify recommendations, or add their own insights. This interaction ensures that human expertise and judgment remain central to the decision-making process while leveraging the analytical capabilities of the AI system.

[0132] In some embodiments, the human review interface also provides access to the underlying data and analysis used by the AI system to generate its recommendations. In some cases, clinicians are able to view the original medical images analyzed by the visual agent, review the literature citations provided by the RAG agent, or examine the conversation logs of the conversation agent. This transparency enhances the interpretability of the AI system's output and allows medical professionals to make more informed decisions.

[0133] In some embodiments, the human review interface includes a feedback mechanism that allows clinicians to provide feedback on the quality and accuracy of AI-generated recommendations. This feedback is used to continuously improve and refine EyeGI's performance over time.

[0134] The combination of the Guardrail agent and the human review interface creates a two-tiered validation process for EyeGI. This approach helps ensure that the system's final output is not only technically sound and compliant with medical guidelines, but also consistent with the judgment of experienced medical professionals.

[0135] EyeGI includes an output module such as Figure 8 As shown in Figure 2, the output module is responsible for generating and delivering final diagnostic and treatment recommendations based on data processed and validated by the multi-agent center, the guardrail agents, and the human review interface. Once the AI-generated recommendations are validated by the guardrail agents and approved by the human review interface, they are sent to the output module for final processing and communication to appropriate recipients, such as patients or other healthcare providers.

[0136] In some embodiments, the output module formats the final recommendations in a standardized structure to ensure consistency and ease of interpretation. The output module compiles information from various sources within EyeGI, including analysis results from specialized agents, validation checks by guardrail agents, and any modifications or approvals made through a human review interface.

[0137] The output module includes a FHIR REST API. In some cases, the FHIR REST API allows seamless integration with other medical information systems and electronic health records. Using the FHIR standard may ensure that EyeGI's output can be easily shared and understood between different medical platforms and providers. Among them, FHIR (Fast Healthcare Interoperability Resources) is a healthcare data standard jointly developed by organizations such as the Health Information and Management Systems Society (HIMSS) and the American Medical Informatics Association (AMIA). The FHIR REST API is based on the FHIR standard and provides a standardized way to access and operate healthcare data. It allows different medical systems, such as electronic medical record systems, clinical decision support systems, medical Internet of Things devices, etc., to interact with data through the HTTP protocol, thereby realizing the sharing and integration of medical data.

[0138] In some embodiments, the output module includes a web-based voice UI. The web-based voice UI provides an interface for delivering diagnostic and treatment recommendations via voice interaction. This functionality enhances accessibility for users who prefer or need voice-based interaction, such as visually impaired patients or healthcare providers who need to access information.

[0139] In some embodiments, the output module also includes an Open EMR EDC interface. The Open EMR EDC interface facilitates direct integration of EyeGI output with electronic medical record systems. This integration allows for automatic updating of patient records with the latest diagnostic findings and treatment recommendations, streamlining the documentation process and ensuring that all relevant information is readily available to healthcare providers.

[0140] In some embodiments, the output module customizes the format and content of the final recommendation based on the intended recipient. For example, recommendations for medical professionals include more technical details and references to medical literature, while output for patients uses simpler language and focuses on practical next steps and treatment instructions.

[0141] In some embodiments, the output module also includes a mechanism for securely transmitting the final recommendation to the appropriate recipient. In some cases, this involves encryption protocols and user authentication measures to ensure that sensitive medical information is protected and only accessible to authorized individuals.

[0142] By integrating these various components and interfaces, the Output Module serves as the final stage of the EyeGI workflow, providing comprehensive, validated, and actionable ophthalmic disease diagnosis and management recommendations.

[0143] In some embodiments, EyeGI also includes a user access control system, such as Figure 10 The user access control system manages the authentication, role assignment, and function access of different user types (such as patients, doctors, and administrators).

[0144] In some preferred embodiments, the user access control system includes an authentication module. The authentication module is responsible for verifying the identity of users attempting to access EyeGI. The authentication module uses various methods to verify user identity, such as a username and password combination, biometric data, or multi-factor authentication technology.

[0145] In some preferred embodiments, the user access control system further comprises a role allocation module. Figure 11 As shown, in some cases, a role assignment module determines the appropriate role for each authenticated user. The role assignment module assigns roles such as patient, doctor, or administrator based on predefined criteria or information stored in the system.

[0146] In some preferred embodiments, the user access control system further comprises a function access controller that manages access to different system functions based on user roles. In some cases, the function access controller includes specialized function modules for different user types.

[0147] In a more preferred embodiment, the function access controller includes a patient function module, a doctor function module, and an administrator function module. Each of these modules controls access to specific functions associated with the corresponding user role. For example, the patient function module allows patients to view their medical records and schedule appointments, while the doctor function module provides access to diagnostic tools and patient management.

[0148] In a further preferred embodiment, the user access control system further includes a logging module. The logging module records system activity and access events, creating an audit trail of user interactions with EyeGI. This logging functionality is useful for security monitoring, compliance purposes, and system performance analysis.

[0149] In some preferred embodiments, Figure 10 As shown, the user access control system also includes a configuration manager. In some cases, the configuration manager allows administrators to modify system settings, update user roles, and manage access permissions. The configuration manager interfaces with the role assignment module to ensure that any changes to user roles or permissions are correctly implemented throughout the system.

[0150] The user access control system provides separate permissions for patient, physician, and administrator roles. In some cases, these role-based permissions extend to the analysis and model update approval process within EyeGI. For example, physicians have access to detailed analysis of patient outcomes and treatment effectiveness, while administrators have the right to approve updates to the AI models used in the system.

[0151] By implementing role-based access control and setting permissions for different user types, the user access control system helps ensure that users can access appropriate functions and information while maintaining the security and integrity of EyeGI. This approach also helps comply with healthcare regulations regarding patient data privacy and access control.

[0152] EyeGI operates through a series of interconnected steps, processing patient data to generate diagnostic recommendations and treatment plans. Figure 12 A flow chart illustrating the sequence of interactions between multiple components in a medical data processing system.

[0153] The process might begin in step 1, where the patient provides symptoms or images to the conversational agent. In step 2, the conversational agent processes the initial information. Subsequently, step 3 involves the conversational agent sending the processed information to the agent hub.

[0154] The Agent Center then routes the data for further analysis. In step 4, the Agent Center sends the data to the Vision Agent for image analysis. Simultaneously, in step 5, the Agent Center sends the data to the RAG Agent for evidence retrieval.

[0155] The Vision Agent performs step 6 to analyze the image and returns the analysis results to the Agent Center. Similarly, the RAG Agent performs step 7 to retrieve relevant medical information and returns this information to the Agent Center.

[0156] In step 8, the Agent Center combines the outputs received from the Vision and RAG Agents. Step 9 involves sending the combined output to the Guardrail Agent for safety checking. The Guardrail Agent performs step 10 to verify the output.

[0157] The validated output proceeds to step 11, where it is reviewed and approved or edited. After review, step 12 involves generating the final recommendation. The process concludes with step 13, where the final recommendation is passed back through the system to the original patient.

[0158] Figure 13 A more detailed sequence diagram is provided to illustrate the workflow of the medical diagnosis system. The sequence begins with step Symptom Image S1, where the patient sends symptoms and images to the Web UI. In step Image Request S2, the Web UI initiates a POST image request to the triage agent.

[0159] Step labeling data S3 involves the triage agent processing this information and forwarding the red flag JSON data to the agent center. After receiving this data, the agent center sends a step classification request S4 to the vision agent and a step evidence request S5 to the RAG agent.

[0160] The Vision Agent responds by returning step score data S6, which includes the diagnostic score and heat map data. Meanwhile, the RAG Agent provides step guidance data S7, which contains guidance snippets.

[0161] The Agent Center then compiles this information into a Step Report Draft S8 and sends it to the Guardrail Agent. The Guardrail Agent forwards the draft for review in the Step Review Draft S9. The clinician then returns the step for edit approval S10.

[0162] Finally, the system transmits the final step suggestion S11 back to the Web UI through the chain and finally reaches the patient.

[0163] Figure 14 A flowchart describing the sequence of medical data processing is illustrated. The process starts with step 1, where patient data is received and processed. In step 2, the embedding of the patient data is stored in a vector database.

[0164] The process may continue with step 3, where the system coordinates the interactions between components. This coordination may extend through multiple parallel paths as shown in steps 4 and 5, where the system processes symptom descriptions and analyzes medical images, respectively.

[0165] After analysis, step 6 might involve retrieving relevant medical information and literature. The process might then proceed to step 7, where the outputs of the specialized agents are combined and sent through multiple pathways for processing.

[0166] In step 8, the system may send the combined output for verification. Step 9 may involve sending the verified output to a human review interface. The process may continue with step 10, reviewing and approving the output.

[0167] Step 11 might show sending the approved output to generate a final diagnosis and treatment recommendation. The process might end with step 12, which might involve interfacing with an EMR system.

[0168] Figure 15 Another sequence diagram shows the interaction flow between multiple components in a medical diagnostic system. The sequence might begin with step Symptom Image S1, where the patient submits a symptom description, medical image, and EHR data to the Web UI. The Web UI might then send the patient data to the Agent Hub in step Image Request S2.

[0169] In step Labeling Data S3, the system might store the patient data as embeddings in a vector database. The Agent Center might then initiate parallel processing flows: step Classifying Request S4 might process symptom descriptions via a conversational agent, step Evidence Request S5 might analyze medical images via a vision agent, and step Scoring Data S6 might retrieve relevant medical information via a RAG agent.

[0170] In step Guide Data S7, each agent may return its processed output to the Agent Hub. The Agent Hub may combine these outputs in step Report Draft S8 to create a comprehensive analysis. Step Review Draft S9 may show the combined output being sent to the Guardrail Agent for verification.

[0171] The validated output may then be sent to a clinician for review in step Edit Approval S10. In a final step Advice S11, the clinician may review and approve the output. Step Diagnosis Advice S12 may show the approved output being sent to the output module.

[0172] Finally, in the final step S13, the system may generate final diagnosis and treatment recommendations. The sequence may end with these final recommendations being presented to the patient via the Web UI.

[0173] In some cases, the workflows illustrated in these figures may demonstrate how EyeGI processes patient data through multiple stages of analysis, validation, and review before generating a final recommendation. This approach may help ensure a thorough evaluation of patient information and maintain human expertise in the diagnostic process.

[0174] EyeGI may be used in a variety of scenarios within ophthalmology, demonstrating its versatility and potential to enhance the eye care process. In some cases, the system may be used for initial patient triage, complex diagnosis, and treatment planning.

[0175] For initial patient triage, EyeGI may process patient-reported symptoms and basic eye exam data. In some cases, patients may enter symptoms such as blurred vision, eye pain, or redness through a web interface or mobile app. A conversational agent may interact with the patient, asking follow-up questions to gather more detailed information about the symptoms and their duration. Simultaneously, the system may analyze any eye images uploaded by the patient.

[0176] A multi-agent hub might then coordinate the analysis of this information. A vision agent might examine any uploaded images for signs of common eye diseases, while a RAG agent might compare reported symptoms with a database of eye diseases. In some cases, a web search agent might retrieve recent research or case studies related to the patient's symptoms. The system might then generate preliminary triage recommendations, suggesting whether the patient needs immediate medical attention, a routine appointment, or home care instructions.

[0177] In complex diagnostic scenarios, EyeGI may assist ophthalmologists in analyzing more complex diagnostic data. For example, in a case of suspected glaucoma, the system may process optical coherence tomography (OCT), visual field test results, and intraocular pressure measurements. The visual agent may analyze the OCT scan for signs of optic nerve damage, while the RAG agent may compare the combined test results with established diagnostic criteria for various types of glaucoma.

[0178] In some cases, the system may identify subtle patterns or correlations in the data that might not be immediately apparent to a human observer. The Guardrail agent may then validate these findings against established medical guidelines and present them to an ophthalmologist through a human review interface. The ophthalmologist may review the system's analysis and, based on their clinical expertise, may request additional information or adjust the diagnosis.

[0179] For treatment planning, EyeGI could assist in developing personalized care strategies. In the case of a patient with diabetic retinopathy, the system could analyze the patient's retinal images, recent blood glucose levels, and medical history. A RAG agent could retrieve the latest treatment guidelines for diabetic retinopathy, while a web search agent could identify any relevant clinical trials or new treatment options.

[0180] Based on this information, the system may generate a recommended treatment plan, which may include medications, lifestyle changes, and recommendations for follow-up appointments. In some cases, the system may also consider factors such as the patient's age, overall health, and previous response to treatment when formulating its recommendations. The ophthalmologist may review and modify this plan through a human review interface before finalizing it for the patient.

[0181] In some cases, EyeGI may also assist in monitoring treatment effectiveness and disease progression. For patients undergoing treatment for age-related macular degeneration, the system might track changes in retinal images, visual acuity measurements, and patient-reported outcomes over multiple visits. A visual agent might analyze serial retinal images for signs of improvement or worsening, while a RAG agent might compare the patient's progress with expected outcomes based on clinical literature.

[0182] The system can generate progress reports and alert the ophthalmologist to any worrying trends or unexpected changes in the patient’s condition. In some cases, this could allow for timely adjustments to the treatment plan or early intervention if the disease appears to be progressing faster than expected.

[0183] In these use cases, EyeGI could potentially maintain a comprehensive record of all analyses, recommendations, and decisions. This information could be securely stored and easily accessed for future reference, potentially improving continuity of care and facilitating long-term patient management.

[0184] By integrating multiple specialized agents and incorporating the latest medical knowledge, EyeGI has the potential to provide valuable support in all aspects of ophthalmic care. However, in all cases, the system is likely to serve as a tool to augment and assist human medical professionals, rather than replace their expertise and judgment.

[0185] EyeGI may offer several potential benefits and advantages in the fields of eye care and ophthalmology. In some cases, a multi-agent AI framework may help improve diagnostic accuracy. By integrating multiple specialized agents, such as a visual agent for image analysis and a RAG agent for evidence retrieval, the system may provide a comprehensive assessment of patient data. This approach may help identify subtle patterns or correlations that might not be immediately apparent to a human observer.

[0186] EyeGI has the potential to improve the efficiency of patient management. In some cases, the system's ability to quickly process and analyze large amounts of data could reduce the time required for initial triage and diagnosis. This could allow healthcare providers to more effectively prioritize cases and allocate resources based on the urgency and complexity of each patient's condition.

[0187] EyeGI has the potential to provide enhanced decision support for healthcare providers. By searching and synthesizing relevant medical literature and guidelines, EyeGI may present ophthalmologists with the latest information to inform their clinical decisions. In some cases, this may help ensure that treatment recommendations are aligned with the latest evidence-based practices in eye care.

[0188] EyeGI has the potential to address certain challenges in eye care delivery. For example, the system's ability to perform initial triage based on patient-reported symptoms and basic eye exam data could help manage patient flow in busy clinical settings. In some cases, this could reduce wait times for patients with urgent conditions while providing appropriate guidance for those with less severe conditions.

[0189] EyeGI's integration of multiple data sources may facilitate a more comprehensive patient assessment. By combining information from medical images, patient history, and current symptoms, the system may provide a holistic view of each patient's eye health. In some cases, this integrated approach may help identify underlying factors or connections that could impact diagnosis and treatment planning.

[0190] EyeGI's use of guardrail agents and a human review interface may enhance the safety and reliability of AI-generated recommendations. This two-tiered validation process may help ensure that the system's outputs are consistent with established medical guidelines and subject to expert oversight. In certain situations, this approach may increase healthcare providers' confidence in using AI-assisted tools for eye care management.

[0191] The system's ability to monitor treatment effectiveness and disease progression over time may support long-term patient care. By tracking changes in diagnostic measures and patient-reported outcomes across multiple visits, EyeGI may assist healthcare providers in identifying trends and adjusting treatment plans as needed. In some cases, this longitudinal analysis may help develop more personalized and adaptive care strategies for patients with chronic eye diseases.

[0192] EyeGI's scalable architecture and use of interchangeable components may allow for flexible deployment and integration with existing healthcare systems. In some cases, this could facilitate adoption across diverse clinical settings, from small clinics to large hospitals, while maintaining consistency in the quality of AI-assisted ophthalmic care.

[0193] By incorporating web search agents to retrieve the latest research and clinical trials, EyeGI may help bridge the gap between cutting-edge research and clinical practice. In some cases, this functionality may allow healthcare providers to consider emerging treatment options or participate in relevant clinical trials, potentially expanding the range of care options available to patients.

[0194] The system’s role-based access control and secure data management capabilities may address issues related to patient privacy and data security in digital health solutions. In some cases, this may help healthcare organizations comply with regulatory requirements while leveraging the benefits of AI-assisted ophthalmic care management.

[0195] While EyeGI may offer these potential benefits and advantages, the system's primary function is likely to augment and support human medical expertise, not replace it. The system may serve as a tool to assist healthcare providers in delivering high-quality eye care, with the ultimate decision-making and patient management strategies remaining in the control of qualified medical professionals.

[0196] Example 3:

[0197] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the multi-agent-based ophthalmic care plan generation method as described in any one of the above embodiments and preferred embodiments thereof are implemented, the method primarily comprising:

[0198] S1. Coordinate multi-agent interactions to process pre-acquired patient data; the multi-agents include at least a conversational agent for patient interaction, a visual agent for image analysis, and a retrieval-enhanced generative agent for accessing the latest medical literature;

[0199] S2. Verify the output result of a single agent in the multi-agent;

[0200] S3. Review and annotate the verified output results;

[0201] S4. Generate the final ophthalmic care plan based on the reviewed and approved output results.

[0202] It can be understood that the electronic device for generating ophthalmic care plans based on a multi-agent provided in the embodiment of the present invention corresponds to the above-mentioned method and system for generating ophthalmic care plans based on a multi-agent. The explanation, examples, beneficial effects and other parts of its relevant contents can refer to the corresponding contents in the method and system for generating ophthalmic care plans based on a multi-agent, and will not be repeated here.

[0203] Example 4:

[0204] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the multi-agent-based ophthalmic care plan generation method as described in any of the above embodiments and preferred embodiments thereof are implemented, the method comprising:

[0205] SS1. Coordinate multi-agent interactions to process pre-acquired patient data; the multi-agents include at least a conversational agent for patient interaction, a visual agent for image analysis, and a retrieval-enhanced generative agent for accessing the latest medical literature;

[0206] S2. Verify the output result of a single agent in the multi-agent;

[0207] S3. Review and annotate the verified output results;

[0208] S4. Generate the final ophthalmic care plan based on the reviewed and approved output results.

[0209] It can be understood that the multi-agent-based ophthalmic care plan generation storage medium provided in the embodiment of the present invention corresponds to the above-mentioned multi-agent-based ophthalmic care plan generation method and system. The explanation, examples, beneficial effects and other parts of its relevant contents can refer to the corresponding contents in the multi-agent-based ophthalmic care plan generation method and system, and will not be repeated here.

[0210] In summary, compared with the existing technology, the present invention has the following beneficial effects:

[0211] 1. The present invention provides a multi-agent-based ophthalmic care plan generation technology. This technology coordinates the interaction of multiple agents, including dialogue agents, visual agents, and retrieval and enhancement generation agents, to process patient data. The technology then verifies the output results of individual agents within the multi-agent system, reviews and annotates the verified output results, and finally generates a final ophthalmic care plan based on the approved output results. This technology incorporates dual input and output protection measures, mandatory manual verification, and role-based access control for patients, physicians, and administrators. Furthermore, the technology is designed for scalability, utilizing interchangeable components and microservers that can be deployed locally or in the cloud. This technology provides an end-to-end workflow for ophthalmic care, improves diagnostic accuracy and data security, and results in better treatment outcomes and enhanced patient management.

[0212] 2. This invention provides a multi-agent-based ophthalmic care plan generation technology that provides an end-to-end workflow within a single platform, covering the entire spectrum of ophthalmic care, from initial triage to diagnosis, treatment planning, and follow-up scheduling. This integrated approach streamlines the process, reduces the need for multiple, disparate systems, and potentially improves efficiency and reduces errors.

[0213] 3. This invention provides a multi-agent-based ophthalmic care plan generation technology. Its EyeGI adopts a multi-layered safety approach, including dual input and output guardrails and mandatory human verification. This design prioritizes patient safety and helps reduce the risks associated with AI-driven medical decision-making.

[0214] 4. This invention provides a multi-agent-based ophthalmic care plan generation technology that offers improved interpretability compared to traditional "black box" AI models. The Vision Agent provides heat maps for image analysis, while the RAG Agent provides inline references for its recommendations, making the decision-making process more transparent and auditable.

[0215] 5. This invention provides a multi-agent-based ophthalmic care plan generation technology. EyeGI's real-time web search capability allows it to incorporate the latest medical literature without the need to frequently retrain the underlying language model. This feature ensures that the system always stays up-to-date with the latest ophthalmic developments.

[0216] 6. This invention provides a multi-agent-based ophthalmic care plan generation technology. The system's scalable and vendor-neutral architecture utilizes interchangeable components and microservers, providing flexibility in deployment and integration with existing healthcare IT infrastructure. This design allows for easier adoption and customization in diverse healthcare environments.

[0217] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0218] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A multi-agent based ophthalmic care plan generation method, characterized in that: The method comprises: Coordinate multi-agent interactions to process pre-acquired patient data; the multi-agents include at least a conversational agent for patient interaction, a visual agent for image analysis, and a retrieval-enhanced generative agent for accessing the latest medical literature; Verifying the output results of a single agent in the multi-agent; Review and annotate the verified output results; Generate final eye care plan based on the reviewed and approved output results.

2. The method according to claim 1, wherein The patient data includes at least symptom description data, medical image data and electronic health record data.

3. The method according to claim 2, wherein The method further comprises: Processing the symptom description data using an audio processor; processing the electronic health record data using a text processor; and processing the medical image data using an image processor; The symptom description data, electronic health record data, medical image data, and metadata related to the symptom description data, electronic health record data, and medical image data are embedded and stored in a vector database.

4. The method according to claim 1, wherein The method further comprises: utilizing a network search agent to retrieve the latest medical literature online.

5. The method according to claim 1, wherein The method further includes: implementing dual input and output protection measures when verifying the output results of a single agent in the multi-agent.

6. The method according to claim 1, wherein When reviewing and annotating the verified output results, the clinician is allowed to edit and output the verified output results through human-computer interaction before the verified output results are output.

7. The method according to claim 1, wherein The method further includes securely transmitting the final eye care plan to an appropriate recipient using encryption protocols and user authentication measures.

8. A multi-agent based ophthalmic care plan generation system, characterized in that: The system comprises: A data processing module configured to coordinate multi-agent interactions to process pre-acquired patient data; the multi-agents include at least a conversational agent for patient interaction, a visual agent for image analysis, and a retrieval-enhanced generative agent for accessing the latest medical literature; an output result verification module, configured to verify the output result of a single agent in the multi-agent; The output result approval module is configured to review and annotate the verified output results; The final output result generating module is configured to generate a final ophthalmic care plan based on the output results after review and approval.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the multi-agent-based ophthalmic care plan generation method according to any one of claims 1 to 7 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multi-agent-based ophthalmic care plan generation method as described in any one of claims 1 to 7 are implemented.

Images