AI“看懂”影像、“理解”病历：“京医千询”大模型以多模态能力推动医疗普惠

Core Insights - The 2025 Medical Value Healthcare Conference and the 6th Taishan Award Ceremony were recently held in Shanghai, focusing on the theme of "AI-driven diagnostic transformation and industry integration" [1] - JD Health showcased its breakthrough achievements in the medical AI field, particularly through its "Jingyi Qianxun" medical model, which emphasizes collaborative innovation and aims to address the challenges of medical accessibility [1] Group 1: Technological Advancements - "Jingyi Qianxun" has achieved comprehensive integration of information from text to imaging, enabling precise extraction of key information from medical records and orders, and constructing complete health profiles [2] - The model demonstrates advanced pixel-level image analysis capabilities, accurately identifying anatomical structures or lesions in medical images, including MRI, endoscopy, and ultrasound [2] - The multi-modal reasoning ability of "Jingyi Qianxun" simulates human diagnostic logic, allowing for reliable diagnostic outputs even from complex medical test results [2] Group 2: Performance Metrics - In authoritative assessments, "Jingyi Qianxun" has shown outstanding performance, achieving 99% in medical calculation ability, 91% in medical Q&A ability, and 91% in text summarization ability, placing it at the forefront of the industry [2] - The model has been validated against a comprehensive evaluation set based on clinical task data from JD Internet Hospital, confirming its leading capabilities in core medical functions [2] Group 3: Future Directions - "Jingyi Qianxun" has evolved from a general base model to a comprehensive model and then to specialized models for specific diseases, reflecting JD Health's long-term commitment to specialized fields [3] - The company plans to continue developing specialized models through technological innovation to enhance medical accessibility, professionalism, and efficiency, aiming to extend quality medical resources to a broader population [3]