Objective This study aims to construct an online triage and registration assistance system based on multi-modal multi-agent collaborative technology. The objective is to address the current inadequacy of intelligence in online hospital triage and registration, thereby enhancing the efficiency of medical resource allocation and improving the patient experience with online appointment scheduling. Methods The system was designed with a hierarchical architecture, comprising an interaction layer, a processing layer (with four collaborating core agents), and a data layer. Four core agents were developed: tool invocation, report parsing, department recommendation, and doctor recommendation. Based respectively on the Doubao series of models and the DeepSeek-R1-671B model, these agents integrate prompt engineering and a Retrieval-Augmented Generation (RAG) architecture to enable multi-modal information processing and precise recommendations. A controlled experiment on triage performance, a NASA-TLX workload assessment for triage nurses, and a patient Net Promoter Score (NPS) survey were conducted at a Grade A Tertiary hospital in Guangzhou. Results The system achieved a triage accuracy of 96.67%, an increase of 12.88 percentage points compared to manual triage (83.79%) (P ＜ 0.01). The accuracy for doctor recommendations was 83.33%. Following the system's implementation, the NASA-TLX workload index for triage nurses decreased by 8.32 (P= 0.03), and the patient Net Promoter Score (NPS) increased from 12% to 44%. Conclusion The multi-modal multi-agent triage and registration assistance system developed in this study successfully facilitates the intelligent transformation of online medical triage. Its triage accuracy surpasses that of human, effectively reduces the workload for triage nurses, and enhances the patient experience with online registration. This system offers a viable technical solution for the advancement of smart hospitals.

Objective To improve the objectivity, consistency, and evaluation efficiency of nursing operation assessments, this paper proposes and implements an intelligent nursing operation assessment system based on multi- modal perception and multi-agent collaboration. Methods Taking clinical nursing operation videos as input, a multi- agent collaborative assessment system was constructed, and the assessment standards are converted into executable tasks. A dual-path reasoning agent was used to analyze operation videos and voice verification information respectively, and then a fusion decision-making agent was employed to complete score calculation and report generation. A verification mechanism was introduced to ensure the reliability and availability of the results. Results In various standard nursing operation scenarios, the system could stably realize automatic evaluation, which had high consistency with expert evaluation in terms of process integrity and operation standardization, and significantly reduced the time cost of manual assessment. Conclusion Organizing and applying multi-modal models in a multi-agent collaborative manner can provide an extensible, interpretable and practically applicable technical path for nursing skill assessment.

Objective To address the challenges of semantic omissions, complex linguistic structures, and terminological heterogeneity in multi-center integrated Chinese-Western medicine electronic medical record (EMR), this study aimed to construct a DeepSeek-based automated normalization framework to efficiently process large- scale heterogeneous data and the associated complex linguistic phenomena. Methods The DeepSeek large language model was employed via its API in combination with modular prompt engineering. The framework performed explicit completion of key semantic components, ambiguity resolution and sentence restructuring, as well as multi- center clinical terminology alignment on raw heterogeneous EMR texts, thereby enabling their transformation into normalized data that were semantically clear, structurally unified, and annotation-friendly. In addition, targeted content validation was conducted in collaboration with medical experts, and the model outputs were subjected to sampling- based review and accuracy evaluation. Results Under zero-shot conditions, DeepSeek-R1 achieved 86% average accuracy in standardizing knee osteoarthritis (KOA) EHRs from three hospitals. The  "LLM-led standardization with human evaluation"  paradigm significantly reduced manual effort while improving efficiency and quality, demonstrating robust capabilities in semantic understanding and reasoning. Conclusion This framework generates high-quality standardized data, establishing a solid foundation for downstream structured annotation and knowledge extraction, and thereby supporting the construction of high-precision, disease-specific knowledge graphs for integrated Chinese- Western medicine.

Objective This study aims to construct a DRG (Diagnosis-Related Groups) full-process management system based on knowledge graphs to address the challenges of refined management in medical institutions posed by DRG payment reform. Methods By integrating clinical pathways with DRG information, a multidimensional medical knowledge graph was constructed using knowledge graph technology. The system includes modules such as pre-admission intelligent decision support, real-time monitoring and intervention during treatment, and closed-loop feedback optimization after discharge, enabling full-process management of DRG. Results After implementation of the system, the admission rate of clinical pathways increased by 24.07%, the optimization rate of pathway versions grew by 17.55%, the variation rate of pathways decreased by 6.49%, the DRG surplus rate rose by 8.03%, and the average length of hospital stay was shortened by 1.02 days. These outcomes indicate that the system effectively integrates medical resources, standardizes treatment processes, improves medical efficiency and quality, and reduces medical costs. Conclusion A DRG full-process management system based on knowledge graphs demonstrates significant effects in enhancing medical efficiency, quality, and economic benefits. It provides strong support for refined management in hospitals under DRG payment reform and has good potential for promotion.

Objective To propose an information security and privacy protection scheme for medical data sharing, to resolve the management, technical, and operational risks faced in the process of medical data sharing, and to achieve comprehensive security protection and privacy preservation for medical data sharing. Methods By enhancing management and leveraging core technologies such as domestic commercial cryptography, data masking, and blockchain, a security protection system is constructed. Additionally, an operational monitoring system based on a closed-loop feedback mechanism is established to ensure the security of medical data sharing and privacy protection. Results We have established a security and privacy protection system for data sharing, covering coordinated governance in security management, technical protection, and operational monitoring. Conclusion This plan introduces the implementation path of information security in medical data sharing   and privacy protection from the perspectives of management, technology, and operational collaborative governance, which has good reference significance for the implementation and security construction of related scenarios.

Objective To develop a large language model-based agent for automated TNM staging of nasopharyngeal carcinoma and to improve the accuracy and consistency of clinical staging. Methods A multi-stage reasoning framework was proposed, comprising information extraction, feature standardization, and guideline mapping. Tool filtering and adaptive context compression were incorporated to improve inference efficiency, and an extensible tool library was established to support staging decisions. The agent operates through a thought–action–observation loop, enabling a transparent and traceable reasoning process. Results In 192 nasopharyngeal carcinoma clinical reports, the agent achieved accuracies of 92%, 74%, and 95% for T, N, and M staging, respectively. Token consumption decreased by 34.2%, 55.4%, and 63.8% under settings with 10, 30, and 50 tools, respectively, with a mean reduction of 51.1%. The agent further generated interpretable reasoning chains, supporting clinician understanding and verification of staging outcomes. Conclusion This large language model-based TNM staging agent offers an effective approach for automating clinical staging, with favorable performance in both accuracy and efficiency. Further validation in multi- center and multi-cancer settings is warranted to determine its generalizability and clinical utility.

Objective To explore the path of establishing a regional digital and intelligent health service system under the interconnection and sharing of national health information, and to improve the quality of regional health services and residents' sense of gain in medical treatment. Methods Build A trusted data space was built based on the computing power base, integrate multi-source heterogeneous health data through data circulation, combine health portrait and artificial intelligence technologies, improve the dimensions of national health information and service collaboration mechanisms, and design a regional digital and intelligent health service system covering the "whole medical process and whole health cycle". Results The system in Nanjing has been connected to  120 medical institutions above the secondary level, with more than 9.5 million registered users and an annual service volume exceeding 20 million person-times; by optimizing the medical service process, it has effectively solved the problem of "three long and one short" (long registration time, long waiting time, long payment time, and short consultation time), achieving significant social benefits. Conclusion The digital and intelligent health service based on the interconnection and sharing of regional smart medical care can effectively integrate medical resources, improve service efficiency, enhance the medical experience, promote the rational allocation of medical resources, realize the management of the whole-life-cycle health information and the whole-process control of medical treatment, and provide practical reference for the construction of Healthy China.

Objective To construct a knowledge graph for the safe use of Traditional Chinese Medicine (TCM) and improve rational clinical medication. Methods Data on Chinese patent medicines were integrated from drug instructions and the Chinese Pharmacopoeia, and an Excel data table was established. This table includes information such as Chinese patent medicines, ingredients, indications, main efficacy, and dosage and administration, and the extracted information was preprocessed. The Label Studio platform was used to create intelligent annotation examples, the API interface of Doubao was called for knowledge extraction, and finally the Neo4j graph database was used for knowledge storage and visual display. Results A TCM safe medication knowledge graph based on Neo4j was constructed, which included 1 702 types of Chinese patent medicines, generating 8 592 entity nodes and 46 409 entity relationships (involving Chinese patent medicines, symptoms, syndromes, etc.). Information query and content visualization could be realized based on the constructed knowledge graph. Conclusion The constructed TCM safe medication knowledge graph can provide auxiliary decision support for clinicians, improve the safety of TCM medication, and serve as a reference for the construction of large-scale Chinese patent medicines knowledge graphs in the future.

Objective The construction of a critical illness big data platform and specialized disease database to assist clinical analysis and utilization of critical illness data. Methods By utilizing big data and artificial intelligence technology, clinical diagnosis and treatment data and equipment monitoring data of critically ill patients are aggregated, and multimodal data is deeply managed and integrated to build a high-quality critical data resource center. Based on this, a sepsis research database is constructed to explore early identification, prediction, and early warning of specialized diseases. Results The critical care big data platform has incorporated the complete historical data of the critical care medicine department, supporting critical care researchers to establish more than 40 research projects, construct sepsis specific disease standard datasets, and improve clinical research levels. Conclusion The critical illness big data platform and specialized disease database are of great significance in reducing the threshold for data usage, improving data utilization efficiency, enhancing the level of critical illness clinical diagnosis and treatment, and promoting the transformation of scientific research achievements.

The resource scheduling problem of hospital information systems (HIS) exhibits NP-hard complexity with multiple constraints and objectives, where traditional scheduling methods struggle to meet real-time and optimization requirements. To address the limitations of classical ant colony optimization (ACO), such as slow convergence and proneness to local optima, an improved ACO with an adaptive pheromone management mechanism is proposed. This algorithm integrates a dynamic evaporation coefficient adjustment strategy driven by resource utilization, constructs a multi-objective Pareto optimal search mechanism, and incorporates multi-dimensional heuristic information tailored to hospital business characteristics. Implemented based on a distributed architecture, it achieves standardized integration with HIS. Experimental results demonstrate that compared with traditional methods, the improved algorithm increases convergence speed by 37.9% and resource utilization by 20.5%, while reducing patient waiting time by 33.6%, verifying its effectiveness and practicality.

Shandong provincial third hospital, in alignment with the national guidelines outlined in the "Notice on Accelerating the Large-Scale Deployment and Application of Internet Protocol Version 6 (IPv6)", has implemented an IPv6-based hospital network planning and cybersecurity protection system. Through a comprehensive analysis of the hospital’s operational requirements and existing network infrastructure, the institution has designed a standardized three- layer IPv6 network architecture. By rationally allocating the IPv6 address space provided by internet service providers (ISPs), the hospital has completed IPv6 network planning and deployment. It has laid a network foundation for the informatization development of the hospital, including the construction of internet hospitals, cloud-based medical films, teleconsultation, and examination and test order entry. To mitigate cybersecurity risks such as network attacks, malware infiltration, and phishing threats, the hospital has developed a multi-layered cybersecurity framework. This system integrates intrusion prevention, antivirus protection, web application security, threat intelligence analysis, and other advanced safeguards. These initiatives aim to drive innovation in healthcare IT, ensure stable network operations, and achieve comprehensive cybersecurity protection. This study provides a useful reference for IPv6 network construction, hospital informatization development, and security protection in the medical industry.

Objective To address the inefficiency faced by clinicians in selecting appropriate referral destinations due to high workload, this study aims to design and validate a referral institution recommendation system based on multi-agent collaboration, integrating a two-tower model and a large language model. Methods Utilizing a multi-agent architecture, the system's functions were encapsulated into four agents: perception & preprocessing, two-tower retrieval, deep reasoning, and resource allocation. The core of the system synergized the efficient vector retrieval capability of a two-tower model with the deep semantic reasoning of the DeepSeek-V3 large language model, enabling a "coarse- to-fine" collaborative recommendation process for candidate institutions. Leveraging 13 245 patient records from a tertiary hospital, a final dataset of 1 155 real-world cases with explicit "discharge by physician's advice for transfer" records was constructed after rigorous screening for model training and validation. Recommendation performance was evaluated using metrics including Hit@K, MRR, and NDCG. System response time and robustness were also tested. Results On the test set comprising 10 medical consortium institutions, the system achieved Hit@1, Hit@3, and Hit@5 of 62.9%, 85.3%, and 93.1%, respectively, with an MRR of 0.714 and an NDCG@3 of 0.653. The average end-to-end response time was 53.2 ms. When 30% of structured test indicators or unstructured text fields were randomly masked, Hit@3 decreased marginally by only 0.9 and 2.9 percentage points, demonstrating strong robustness. Conclusion The proposed system delivers accurate, interpretable, and rapid recommendations for referral institutions, effectively narrowing clinicians' decision-making scope and offering a viable technological approach to alleviate the burden associated with the referral process.

Objective To address the code management challenges brought about by the increasing complexity of hospital software systems, and to solve the problems of missing version control, low efficiency in collaborative development, and potential security risks in code storage under the traditional model, a hospital software code management platform that meets the requirements of information innovation should be constructed to achieve standardized management of source code. Methods By analyzing the current state of hospital code management and the adaptation needs for information innovation, a platform architecture centered on containerization technology is proposed. The platform integrates four major functional modules: code hosting, version control, collaborative development, and security auditing. It utilizes domestically produced information innovation ecosystem components to achieve software and hardware adaptation, and ensures the security of the code throughout its lifecycle through a multi-layered security protection mechanism. Results After the platform was deployed in a hospital, the code version conflict rate decreased by 68%, collaborative development efficiency improved by 55%, and the compliance rate of core business system code reached 100%. The security protection system successfully prevented the risks of source code leakage and tampering, demonstrating its technical feasibility and practicality. Conclusion The code management platform based on the information innovation environment effectively addresses the issues of standardization, security, and collaboration in code management during the hospital's informatization process, providing an independently controllable solution for the medical industry. This platform not only significantly enhances the efficiency of code management and the ability to ensure information security, but also offers reusable practical experience for the industry's information innovation transformation and sustainable development.

With the application of emerging information technologies in the healthcare industry, traditional information project management models can no longer meet the needs of modern smart healthcare development. As an important means to improve the quality and efficiency of information system construction, the urgency of building a healthcare information project management system has become increasingly evident. This paper achieves standardized management of the entire process of citywide healthcare information projects through the project management system, formulates and issues corresponding information project management measures, and ultimately realizes full life cycle management from project initiation and application to acceptance. The application of the system has improved the effectiveness of information project management across medical and health institutions in the city and enhanced overall management levels. At the same time, the system has established a project asset database to assist health administrative departments in managing information construction projects.

The simulated court teaching in medical colleges faces challenges such as outdated cases, insufficient authenticity of scenes, and a single evaluation system. It cannot meet the personalized and diversified needs of students, and also cannot meet the demand for health law talents in the era of artificial intelligence. Artificial intelligence empowers medical colleges with effective measures such as creating a dynamic digital case library, accurately and efficiently retrieving legal literature, assisting in case analysis, providing virtual simulated court exercises, and constructing a diversified evaluation system. These measures effectively enhance the effectiveness of simulating real courts, improve the teaching efficiency of simulated courts, and cultivate students' abilities in solving legal problems and critical thinking.