The evolution of artificial intelligence in the healthcare sector is shifting focus from raw model capabilities to the development of robust, system-level applications that can be integrated into real-world clinical workflows. This transition was a key topic at the recent 8th Beijing Zhiyuan Conference, where Dr. Li Linfeng, Vice President of Technology Innovation and AI Architect at YIDU TECH (02158), delivered a presentation outlining the technical framework and clinical application pathways for medical intelligent agents.
The Next Phase of Medical AI: Intelligent Agents as a Key Breakthrough
With a series of "AI+" policies being introduced at the national level, there is a clear directive to promote the deep application of AI in areas such as assisted diagnosis, health management, and chronic disease services. On the technological front, intelligent agents possessing memory, planning, and tool-calling capabilities are emerging as a core direction for the next generation of AI. Dr. Li pointed out that medical scenarios are inherently suited for an intelligent agent framework. Clinical diagnosis and treatment constitute a continuous decision-making process, from pre-diagnosis triage to in-treatment care and post-diagnosis management, featuring clear, step-by-step workflows that align perfectly with an agent's task decomposition and execution logic. Furthermore, the medical industry has stringent requirements for explainability, controllability, traceability, and stability of outcomes, which traditional large language model "black box" interactions struggle to meet for clinical safety and regulatory needs. The complexity of hospital systems and the heavy administrative burden on medical staff also create a pressing need for AI solutions that can interface with various tools and automate high-frequency tasks.
A Two-Platform, One-System Framework: Building the Engineering Foundation for Medical Agents
To address the key challenges of moving intelligent agents from capability to systematic deployment, YIDU TECH has introduced a "two-platform, one-system" architecture covering the entire lifecycle of development, application, and evaluation. The Intelligent Agent Development Platform modularly packages large model capabilities, medical knowledge bases, and tool components, enabling the rapid construction of specialized clinical scenario agents. To date, the company has collaborated with several leading tertiary hospitals and clinical experts to develop over 280 intelligent agents covering core scenarios like assisted diagnosis, treatment plan recommendations, and medical record generation. The Intelligent Agent Application Platform deeply integrates with core hospital systems like HIS and EMR, embedding agents into physician workstations. This enables automatic patient data access and dynamic agent scheduling, allowing AI to be natively integrated into clinical workflows. The Intelligent Agent Evaluation System establishes a multi-dimensional assessment mechanism focusing on trustworthiness, safety, efficiency gains, workflow integration, and physician acceptance, continuously validating the agents' usability and stability in real-world environments. Dr. Li emphasized that the optimal technical pathway follows a "workflow-driven specialized disease scenario agent + multi-agent collaboration" architecture. This model decomposes complex medical tasks into structured workflows, ensuring each step's output is explainable, controllable, and traceable, while integrating guidelines, knowledge graphs, and specialized disease models to significantly enhance stability and consistency in clinical settings.
From an Agent Matrix to Clinical Closure: The Practical Implementation of Dr.Copilot and Yidu Zhi Xun
To make intelligent agents genuinely serve physicians in their daily work, YIDU TECH launched the clinical assistant Dr.Copilot. This system embeds as a lightweight plugin in the physician workstation, allowing agents to be summoned with one click. It automatically calls upon both structured and unstructured patient data to perform reasoning and execute tasks. As the number of specialized and disease-specific agents grows, the platform provides a unified interaction portal. Physicians can input requests in natural language, and the system automatically identifies the intent and matches it with the appropriate agent. Concurrently, the company has introduced a "Task Planning Engine" that transforms standardized diagnosis and treatment pathways into executable sequences of intelligent tasks. For instance, for a hospitalized patient's examination, treatment, and follow-up plans, the system can pre-schedule and run them in the background, with the physician only needing to confirm or adjust, significantly reducing repetitive workload. To extend capabilities to scenarios outside the hospital, YIDU TECH also launched the mobile evidence-based decision support tool "Yidu Zhi Xun." It provides real-time assisted decision-making based on a trusted medical evidence system. Its knowledge base aggregates over 20 million Chinese and English medical literature articles, more than 50,000 clinical guidelines and expert consensuses, and over 60,000 types of medical science knowledge, integrated with specialized knowledge bases from People's Medical Publishing House. This enables "full-text evidence-based direct answering, with every statement traceable to a source," ensuring each conclusion can be traced back to its original reference, serving as a physician's "second brain" and a multi-disciplinary intelligent agent advisory panel. Dr. Li stated, "What truly creates value in medical scenarios is not a single model, but a system of intelligent agents that can be embedded into real workflows, stably execute tasks, and continuously evolve." Currently, YIDU TECH is leveraging its evidence-based, traceable medical intelligent agent system to upgrade AI from an auxiliary tool to clinical infrastructure, pushing it into the core of medical decision-making chains and workflows. This effort aims to advance medical AI from "capability demonstration" to "system-level productivity," accelerating the large-scale and equitable implementation of precision medical capabilities.
Comments