Large language modeling (LLM) has powerful learning and reasoning ability, but due to the lack of internal knowledge, the model faces hallucination challenges. To address the need for high-quality professional knowledge in medical Q&A, this study proposes a medical Q&A method by integrating LLM with knowledge graph, and applies it to Q&A system construction of Medical Science and Technology Information Knowledge Service Platform (MedKaaS). The study shows that compared with the traditional knowledge base Q&A (KBQA) and the emerging Q&A based on generative LLM, the method of integrating LLM with knowledge graph proposed in this paper can fully synthesize the advantages of both, provide evidence traceability support for medical Q&A, and effectively improve the quality of medical intelligent Q&A and the effect of medical knowledge service.

Objective To provide intelligent and convenient medical services for patients and provide more efficient and precise working methods for doctors by building an AI-based pre-consultation system, effectively utilizing patients' pre-consultation waiting time, simulating doctors' consultation process, collecting patients' medical record in advance, and automatically generating structured medical records. Methods By integrating natural language processing and machine learning technologies, the automatic medical records were analyzed in-depth to achieve precise disease analysis and intelligent medical services. By introducing voice, image and other input methods to enrich the means of pre-consultation information collection and improve the accuracy of diagnosis. Results The working efficiency of outpatient doctors was increased since the writing time of medical records was shortened by 34.44%. Conclusion The construction results of the AI-based pre-consultation system can enrich the connotation of medical services, and provide a practical application scenario and foundation for constructing an intelligent, humanized and personalized medical service system.

Objective IT application in hospital operation and management is a key initiative to enhance the comprehensive strength and competitiveness of hospitals. It can not only improve the management efficiency and decision-making level of hospitals, but also improve the service quality and patient satisfaction, promote the rational allocation and utilization of medical resources, as well as promote the digital transformation and intelligent upgrading of the entire healthcare industry. Methods Based on 247 Chinese literatures in CNKI database from 2004 to 2024, research hotspots, sources and evolution trends in this field were extracted by using CiteSpace V6.3 R1 (64-bit), CNKI visual analysis software and online bibliometric analysis platform. Results From 2004 to 2024, the quantity of research papers on IT application in hospital operation and management showed a rising trend year by year. Performance management is the core content of hospital operation and management (including the DRG performance and "Guokao", which means the national accreditation for hospital levels). The pursuit of high-quality development in public hospitals and developing IT application as an important means of hospital operation management modernization are current research hotspots. The research trends include: application of performance evaluation to improve operational efficiency, IT application in hospital operation management to facilitate internal control management and empower hospitals' high-quality development. Conclusion To promote IT application in hospital operation and management in China, efforts should be made in strengthening policy support, increasing investment, promoting standardization, cultivating professional talents and focusing on information security protection.

This paper studies the development history of "Internet+" medical service at home and abroad, applies VOSviewer software to explore the research hotspots and trends in this field, summarizes the current main application scenarios of "Internet+" medical service and its effects, and summarizes the challenges that "Internet+" medical service may face at present or in the future. On this basis, a series of development goals and strategies for "Internet +" medical applications are proposed, which cover online diagnosis and treatment, smart medical and nursing services, primary medical services, medical insurance services, health and disease prevention services, etc., aiming to provide useful references for the development of health services in the new era.

An electronic medical record (EMR) auto-generation system has been established in Sun Yat-sen University Cancer Center based on medical large language models, via data collection, model training and system integration. It has initially realized the automatic generation of discharge summaries, differential diagnosis and progress records, which improved the efficiency of clinical documentation for physicians. Drawing upon the practical experience of this system development, this paper discusses the design and construction of an EMR auto-generation system from the perspective of practical operation, and delves into how to better apply large language models in the field of EMR generation.

Objective To explore the innovative strategy of "Internet+" nursing operation model, and optimize the nursing service experience of discharged patients and special people with mobility difficulties. Methods The Internet and information technology were used to integrate online and offline business, and the  "Internet+" nursing platform was built with the carrier of Wechat official accounts and Wechat Applets and with the technical solutions of separating the front-end and back-end. Results Among the 9 nursing services carried out from July 2021 to July 2023, the number of service cases in 2023 increased by 233.40% compared with that in 2022. At the same time, patient satisfaction increased from 97.52% to 98.42%. Conclusion The "Internet+" nursing platform can ensure the effective integration of online and offline nursing business processes, realize data sharing, process optimization and service mode transformation, enable more patients to enjoy the convenience and quality nursing services of "Internet+" nursing.

Objective To study the framework system of regional large-scale and multi-modal population-based cohort construction, so as to provide reference for constructing a more standardized and a higher level large-scale and multi-modal population-based cohort. Methods Through literature research, case analysis and other methods, the status quo of large-scale population-based cohort construction at home and abroad was deeply analyzed, the challenges in the construction process were systematically sorted out, and a framework system suitable for cohort construction in China. Results The framework system thoroughly sorted out the standardization path of population-based cohort construction and the standard specification system to be followed,  proposed a digital capability support system throughout the whole process of population-based cohort construction, integrated digital technology with  population based cohort characteristics, and provided theoretical and practical guidance for building a more standardized and higher level population-based cohort. Conclusion A set of cohort construction framework system in consistent with the characteristics of Chinese population has been formed in order to improve the construction and operation of Chinese population cohort and promote the research of related disciplines.

Objective To explore the construction of VBHC (Value-based Healthcare)-oriented operational data center in large public hospitals, so as to facilitate medical service model innovation and promote the high-quality development of public hospitals. Methods Under the connotation of VBHC, the construction mode of multi dimensional, multi-level and multi-modal operational data center was constructed from the aspects of healthcare quality, medical insurance payment regulation, business-financial integration, patient service, fine management of logistics  and medical engineering. Results A VBHC-oriented operational data center was established, which facilitated the construction of the hospital's operational management system. Conclusion The concept of value-based healthcare can provide effective guidance for hospitals to provide high-quality medical services, control operating costs and maintain sustainable development. The operational data center based on this construction can provide index data reference for the fine management of hospitals.

Objective To integrate the domain knowledge of Traditional Chinese Medicine (TCM), take the auxiliary diagnosis and treatment as the application  orientation, build the whole domain ontology and knowledge base of TCM, realize the visualization and query application research of TCM knowledge graph, and provide the structure and method basis for building an intelligent reasoning system with dialectical thinking. Methods Taking TCM teaching materials and other domain knowledge as knowledge source, adopting the seven-step domain ontology construction method and based on network ontology language and Protégé ontology modeling tool, the whole-domain ontology of TCM was built. MySQL database was used to store knowledge base data, and Neo4j was used to build the knowledge graph for application research. Results The TCM ontology and knowledge base comprehensively showed the knowledge system of TCM treatment based on syndrome differentiation. There are 17 entity classes, 11 object attributes, 113 data attributes, and 2,012 cases. The knowledge graph contained 3,464 entity nodes and 5,164 relationships, realizing the query and visualization functions of the knowledge graph. Conclusion The construction and application of TCM ontology knowledge base and knowledge graph have certain reference and application value for knowledge representation, sharing and reasoning of TCM diagnosis and treatment experience.

Based on the current situation of financial management in large group hospitals, a unified payment platform applicable to the management mode of multi-campus hospital was designed. The platform covered five functional modules, including transaction order center, unified refund management, automatic reconciliation management, financial statement management and system management. On the basis of the unified payment platform, the hospital smart financial management system was constructed to realize one-stop and convenient settlement services for patients' medical treatment and the integrated and smart financial management under the multi-campus mode. The application of the unified payment platform can effectively optimize the financial settlement process, improve the patient's medical experience, and lay a solid foundation for comprehensively promoting the smart financial management of hospitals.

Objective To assist the monitoring,treatment and research of common geriatric diseases through applying intelligent technology,so as to solve the geriatric medical problems brought about by the accelerated aging population.Methods With the help of medical Internet of Things(IoT)smart devices,the necessary out-of-hospital health data of the elderly was monitored and integrated in real time,and a closed-loop health data collection for common geriatric diseases was formed.An integrated cloud monitoring platform for common geriatric diseases and a multidimensional health data center platform was established.Based on big data governance technology and AI technology,the research of clinical intelligent recommendation decision-making model for geriatric diseases was carried out.Results The integrated cloud monitoring platform has been applied in some hospitals on a pilot basis,while the multi-dimensional health data center platform for common geriatric diseases has been initially built,effectively making up for the shortcomings of the previous geriatric health data center.Using more than 220,000 geriatric health data and hospital clinical knowledge base,a knowledge map of common geriatric diseases was constructed and a clinical intelligent recommendation decision-making rule engine was initially formed.Based on the knowledge map,the intelligent rule engine and intelligent recommendation decision-making model,clinical intelligent recommendation decision-making support for common geriatric diseases were realized with the help of the integrated application of CDSS.Conclusion This study has a positive impact on promoting regional geriatrics IT application,hierarchical diagnosis and treatment,clinical decision-making support and chronic disease management.

For thyroid nodules, traditional imaging diagnosis is usually based on the image observed by the doctor's naked eyes, and the result interpretation is subject to the subjective influence of the observer, and the deep information hidden in the image cannot be recognized. Radiomics is a new non-invasive technology, which transforms ordinary image into quantitative data features through high-throughput feature extraction algorithms, so as to reflect the underlying physiological and pathological features of lesions. With the application of artificial intelligence (AI) technology, radiomics has become a hot spot of clinical application research, and certain achievements have been made in the identification of benign and malignant thyroid nodules and the prediction of cervical lymph node metastasis, which can provide important information for the preoperative decision-making of thyroid nodules. This paper reviews the research progress of ultrasound, CT, MRI and PET/CT multi-domain imaging integrated with AI technology in the clinical application of thyroid nodules.

Objective To analyze the research status, research topics and development trends of large language models in the medical field. Methods A bibliometric analysis was conducted on 518 relevant literatures retrieved from PubMed database. The Latent Dirichlet Allocation model was applied for topic modeling, Python was used as the programming language for the bibliometric analysis algorithm, and Gephi was employed for visual analysis. Results The trend of the number of publications indicates that the research on large language models has increased steadily and rapidly in recent years. The most prolific journal is the Journal of Medical Internet Research, with the United States leading the field by virtue of the number of articles published, and the University of  California is the institution with the largest number of publications. In addition, Geoff Hollis and Siru Liu are prolific authors in the field. Conclusion The research on large language models has attracted much attention in the medical field. Large language models not only provide new tools for clinical assistant decision making, medical records management, and drug development, but also facilitate improvements in intelligent assistant, medical education, and patient communication. However, there are still many challenges faced in their development and application.

Objective To establish a digital twin capability model for clinical research under data-intelligence integration, promote the digitalization, virtualization and visualization of the whole-process elements of clinical research and further strengthen the digital twin service management capability in medical and pharmaceutical fields. Methods By systematically sorting out the development status quo of digital twin in the medical field, and based on the actual development demands of clinical research, the evolutionary path of digital twin in clinical research was designed at the top level, and the feasibility of the proposed model and technical route were verified through pilot application. Results The digital twin capability model for clinical research was constructed under data-intelligence integration, and the conceptual category, stage goals and the evolutionary path formed with the goals of each stage of digital twin were clarified, which helped to create a new model, new standard and new ecology of clinical research from the perspective of twin technology. Conclusion The construction of digital twin capability model for clinical research, to some extent, can help hospitals at different stages of development to clarify the construction path of digital twin capability for clinical research, so as to avoid deviating from the actual service demands and application scenarios of clinical research, and ensure the consistence of goal-oriented and demand-orientated development.

Objective To explore the construction and application of clinical research data platforms. Methods Taking the First Hospital affiliated to University of Science and Technology of China (Anhui Provincial Hospital) as an example, a technical framework of clinical research data platform was constructed, a four-tier data system of operational data storage, clinical data center, research data center, and specialized disease database was established. Data collection and integration of hospital information system were carried out, and data quality control, data labeling, and data security management were carried out to establish a high-quality data platform that can support clinical research. Results The clinical research data platform achieved independent and streamlined clinical research services such as medical data retrieval, specialized disease management and data analysis. Conclusion Clinical research data platform can deposit data assets, meet the data usage needs of medical research, so as to promote the transformation of data from passive support to active intelligence, and stimulate data productivity.

In order to deeply explore the methods of improving the efficiency and quality of orthodontic diagnosis and treatment, and to address the problems of insufficient professionalism, uneven distribution of medical resources and weak timeliness of treatment in the current orthodontic treatment work, advanced technologies of artificial intelligence (AI), big data, and cloud computing were adopted to innovatively develop an orthodontic remote diagnostic and treatment platform. Its architecture and main functional modules were designed to provide intelligent diagnosis and analysis, treatment plan suggestions and remote diagnosis and treatment services. A primary medical institution was selected to put the AI + orthodontic remote diagnosis and treatment system into practice, and its effect was evaluated. The application results show that after applying this system, the accuracy of the treatment plan and patient satisfaction are significantly improved, and the work difficulty of medical staff is also significantly reduced, which fully verifies the practicability and important value of the system design.

This study aims to examine the development status and application prospects of urban pre-hospital first aid information system. Pre-hospital first aid is an important part of urban medical rescue operations, and enhancing the capacity of IT application is of great significance to improve the efficiency and quality of first-aid services. China has implemented a number of measures to improve the level of IT application in first aid services, and established a standardized pre-hospital network layout and IT application framework. This research proposes a construction plan of the urban pre-hospital first aid information system based on new technology applications, including 5G medical emergency private network, emergency operations center and accurate scheduling system. The proposed system intends to enhance the response speed and quality of emergency services by means of real-time data sharing, accurate positioning and scheduling, and remote medical consultations, thereby realizing the seamless integration of pre-hospital and in-hospital information, which enables the critical information from emergency scenes and patients' vital signs be transmitted to hospitals in real time, so as to provide scientific, precise decision-making support for medical professionals and dispatchers. By discussing the pre-hospital first aid information system, this study offers a scientific basis and technical  path for enhancing the quality and efficiency of pre-hospital fir Pre-hospital first aid services in urban areas of China.

In order to solve the problems of tediousness and inefficiency in medical affairs management, a medical affairs management system was built based on the hierarchical evaluation standards of hospital intelligent management, which realized the digital management of physician HR files, qualification  authorization, medical security, etc., optimized the medical affairs management process, improved the quality and efficiency of medical affairs  management, and provided strong support for the construction of smart hospitals.

Objective To enhance the informatization and intelligence of maternal and child health management.Methods Artificial intelligence, real-time data sharing and exchange, mobile health management and other technologies were used to establish a smart maternal and child health information management model, and a smart maternal and child health management information platform was built, which realizes vital signs collection, health monitoring, early warning intervention and follow-up management throughout pregnancy, childbirth and postpartum rehabilitation and other stages. Results Under the whole-process intelligent management model, the time required for maternal and child healthcare services was significantly reduced, the completeness and accuracy of medical records were significantly improved, maternal satisfaction and overall experience in healthcare services were also significantly enhanced (P＜0.05). Conclusion The smart maternal and child health information management platform has realized the informatization of the whole process of  maternal and child healthcare services, and improved the quality of medical services in hospital.

Objective To establish hospital API interface asset ledger, realize unified and centralized management of API interfaces, and monitor the operational state of API interface in real time, so as to discover and rectify interface security risks in time, improve the level of information security protection in hospitals. Methods Taking a tertiary Grade A public hospital as an example, the hospital API interfaces were sorted out and a ledger was established by integrating manual and systematic identification. By utilizing the API interface auditing equipment to monitor the network traffic in real time, identify security risks and rectify and strengthen them. Results The centralized and unified management of API interfaces in the whole hospital was realized. Through the traffic analysis of API auditing equipment, the hidden security risks existing in API interfaces were identified, rectified and strengthened, which improved the level of information security protection in the hospital and ensured the safe sharing and circulation of medical data among information systems. Conclusion The security monitoring of API interfaces is helpful to identify the security risks of application systems in medical institutions, improve their network security protection capability, and safeguard the security of medical data.

Objective To realize the whole-process and closed-loop management of medical safety (adverse)events, enhance the initiative of reporting, and standardize and efficiently dispose of adverse events. Methods By sorting out the adverse events management process and system, designing and optimizing functional modules such as event reporting, department assignment, analysis and rectification, supervision feedback and statistical analysis, and completing code writing, the information management of medical safety (adverse) events was realized. Results After the operation of the system, the reporting rate of adverse events and staff participation were significantly improved, and all dimensions were better than the average of similar hospitals in our country, and the whole-process and closed loop management of adverse events was realized. Conclusion The use of medical safety (adverse) events information management system can realize the information management of adverse events, improve the initiative of reporting and the management level of incident disposal, and provide strong support for the quality and safety management of hospitals.

Objective To analyze the status quo, hotspots and trends of AI research in the field of emergency care.Methods A search in the core collection of Web of Science database was conducted for literature on AI applications in emergency care, which was visualized with CiteSpace software. Results A total of 1,685 documents were obtained, showing a significant upward trend in publications began in 2018, with the United States leading in publication volume (692) and centrality (0.44), followed by China (214) in volume. Harvard University was the highest-ranking institution in publication volume and centrality. The research hotspots of AI application in the field of emergency care mainly focuses on predicting emergency admission rate, developing models for disease prediction and diagnosis, assisting triage in clinical decision-making system, and optimizing resource allocation. Simulation optimization is the future research trend of AI application in emergency care. Conclusion The application of AI in the field of emergency care has great potential, but also faces challenges. Future research should focus on the solution of existing problems and the development of new technologies on this basis, so as to further promote AI  application in the field of emergency care.

Objective To train a deep learning model based on the combination of convolutional neural network (CNN) and bi-directional long and short-term memory (BiLSTM) network on electrocardiogram (ECG) signal dataset to achieve efficient classification of ECG signals. Methods The CNN+BiLSTM model was proposed to classify ECG signal dataset, and CNN was used to extract the initial features of ECG signals. These features were learned more deeply through a BiLSTM network to capture the temporal information and spatial correlation in the signal. Results Through comparative experiment, the CNN and BiLSTM network were effective in improving the model performance. The classification accuracy of the model was 98.91%, the recall rate was 98.71%, the accuracy was 99.79%, and the F1 index was 99.14%. Conclusion Integrating and applying CNN and BiLSTM network to classify ECG signal dataset can effectively improve the flexibility of clinical diagnosis, and provide an effective reference for the automatic analysis of ECG signals.

Objective To construct a full-scene "Internet+" emergency treatment system based on 5G and AI technology, extend the in-hospital service capacity and resources to out-of-hospital through the Internet and intelligent means, and solve the current challenges of emergency medical services, such as difficult data collection, unstable network transmission, insufficient collaborative interaction and insufficient intelligence of auxiliary diagnosis, aiming to construct a patient-centered hierarchical intelligent emergency treatment system to realize the integrated remote emergency medical services inside and outside the hospital. Methods In the event of an emergency, the "voiceprint +" identity authentication technology was used to ensure the trusted access of frontline rescuers, emergency doctors, experts and patients. The 5G technology was used to realize multi-party telemedicine collaboration. Based on remote real-time monitoring, remote consultation and AI-assisted decision-making, the intelligent generation of emergency response plans, intelligent disease monitoring and early warning and hospital and route intelligent planning, in order to  remotely guide the emergency treatment and make the treatment plan in advance. Results The system has been applied in many units such as Tsinghua Changgeng Hospital, Zhuhai People’s Hospital, Fujian Provincial Emergency Center, and has successfully supported the emergency treatment for many major events such as the World Internet Conference and Beijing Winter Olympics, and has served millions of residents in the Tiantongyuan community. Conclusion By integrating intelligent technology and network technology, moving forward the gate of emergency treatment and the capacity of emergency treatment, an intelligent treatment system based on 5G technology and AI technology with the front-end first-aiders providing emergency treatment and the back-end hospital specialists participating in-depth is established, forming the management specification and the technical standards of the remote intelligent emergency treatment system, creating a new model of emergency treatment management with full regional coverage, all-disease service and whole-process management.

Objective To develop an intelligent medical record coding model integrating deep learning and knowledge graph technologies, so as to solve the problems such as low quality and efficiency of traditional manual medical record coding methods. Methods First of all, Natural Language Processing (NLP) technology was employed to develop an electronic medical record information extraction model, which can extract text feature vectors related to electronic medical record coding and identify clinical diagnosis entities. Secondly, a text multi-label classification model was designed to classify and code clinical diagnosis entities, and obtain preliminary ICD codes. Then, a core term matching model based on the knowledge graph was constructed to further select the preliminary coding results. Finally, two comparative experiments were designed and implemented to compare the coding performance of the intelligent medical record coding model with traditional methods and conventional manual coding. Results The coding performance of the intelligent medical record coding model developed in this study was better than that of the convolutional neutral network (CNN) model, the BiLSTM+Attention mechanism model and the intelligent medical record coding model without simple data disturbance enhancement presented herein. The accuracies of principal diagnosis coding and principal operation coding concerning malignant tumors from the intelligent medical record  coding model were 96.6% and 98.9% respectively, which were significantly higher than the accuracies of the traditional manual coding (85.3% and 89.4%). Conclusion The intelligent medical record coding model can enhance the quality of medical record coding while improving coding efficiency, which can provide higher quality data support for subsequent DRG/DIP pre-grouping and fine operational management.

Objective To improve the traditional performance appraisal methods, such as addressing the problems of unclear appraisal indicators and unstandardized data collection and analysis. Methods Using technologies such as AI and natural language processing (NLP), the scattered data of various systems were integrated and managed in a unified manner, stored in the data center in the form of reported indicators, so as to achieve the universal definition of indicators, build a performance appraisal system, display and analyze the trend of related indicators, and dynamically track and manage them. Results The platform integrated functions of display, statistics, analysis and feedback, achieved intelligent indicator analysis, fine operation management, dynamic performance monitoring and early warning, and improved the data quality. Conclusion The application of this system improves the accuracy and efficiency of hospital performance appraisal, and can provide decision-making support for hospital management.

As an important part of the operation and management of public hospitals, the fine management of human resources can not only improve the comprehensive quality and competitiveness of employees, but also improve the overall operational efficiency and service quality of hospitals. Starting from the construction of medical resource center focusing on the core business of medical staff business, position staffing, department structure optimization and professional structure optimization, this paper introduces the overall positioning and index analysis design of the medical human resources platform, and elaborates on the core functions of the system, such as model management, assessment plan management, and evaluation result management. Finally, the application practice is presented from the system level. Supported by data, the overall project can promote sustainable development of disciplines, rational utilization of personnel, efficient operation of business, high-quality development and digital transformation of hospitals.

How to improve the quality and efficiency of medical services by IT application is a realistic issue faced by hospitals at all levels. Based on the hospital's actual situation, this study sorted out and classified the problems existing in the operation of the hospital under the DIP (Diagnosis-Intervention Packet) payment mode, explored the application of information technology to achieve fine operation and management through advance reminder, in process supervision and post-analysis, further analyzed the potential problems of information support, proposed countermeasures and suggestions on the basis of the results achieved in the analysis.

Objective To use information technology to build a "one-stop" smart nursing work platform. Methods A knowledge base of various types of rules was built around clinical nursing work, and driven by the "reminder scheduling" engine, the decentralized nursing system was integrated to build a "one-stop" smart nursing work platform.Results The "one-stop" processing was realized in the clinical nursing platform, which supported the reminder and inquiry of nursing work, and enabling quality control and verification of the completion of nursing work. Conclusion The construction and application of the platform can integrated information technology into clinical nursing work more comprehensively, solve the problem of nurses' frequent operation among multiple nursing business systems, and significantly improve the work efficiency of nurses.

Objective To optimize the process of submitting pathological specimens in the operating room, improve the quality and efficiency of submitting specimens, standardize the work behavior of medical and nursing staff, and meet the requirements of tertiary hospital qualification review. Methods Adopting the   moile nursing system model, mobile PDA device and thermal paper printer, the electronic, standardized and efficient specimen information was realized. At

the same time, persons in charge of each link were collectively trained to realize standardized operation. Results The optimized process of pathological  specimen delivery for inspection in operating room realized the automatic collection and processing of information, reduced the time of information   verification, reduced the problems of specimen omission and retention, reduced the nursing errors, clarified the responsibilities of each link, and reduced the error rate of delivery for inspection. Conclusion The electronic specimen information can make the process of hospital pathological specimen delivery for inspection electronic, standardized and efficient, which is of great significance to improve the hospital management level and patient safety.

While bringing revolutionary changes to the medical industry, artificial intelligence (AI) also brought ethical, legal and social challenges, which has posed a severe test to AI governance frameworks around the world, and spawned an urgent need for new governance tools. With its unique adaptability and flexibility, soft law governance has become an important means to address the legal challenges faced medical by AI intelligence. Soft law has obvious advantages in responding to technological development, meeting social needs, and adapting to cultural diversity, but it also faces practical challenges such as insufficient refinement, difficulty in balancing interests, and poor implementation effects. In the future, China should further strengthen the advantages of soft law governance in medical AI, overcome its shortcomings. By enhancing the effective convergence between soft law and hard law, strengthening global cooperation in soft law, accelerating the localized and integration of soft law, and improving the transparency and public participation of soft law, the implementation effect of soft law can be improved, so as to promote the healthy development of medical AI.