Beijing Friendship Hospital Deploys DeepSeek for Medical Imaging Reports and Knowledge Q&A

Title

Beijing Friendship Hospital Deploys DeepSeek for Medical Imaging Reports and Knowledge Q&A

Content

Beijing Friendship Hospital, a tertiary general hospital in Beijing, deployed a DeepSeek-based AI system for two application scenarios: medical imaging report generation and hospital knowledge-base intelligent Q&A. The imaging report system covers all campuses of the hospital's imaging center with more than 330 installed devices, with physicians using it about 10,000 times per month, and radiates to regional medical institutions through the Beijing Medical Imaging Quality Control Center, improving imaging report efficiency and quality. The project began in October 2022 with literature review, expert interviews and field research, collecting nearly five years of historical imaging examination data to build an experimental corpus. From August to November 2023, the team sampled 10,000 historical imaging reports dating back to 2015 from the hospital's PACS/RIS, structured the data with medical text cleaning and desensitization tools, and built a 200-million-token imaging report corpus. Incremental pre-training and instruction fine-tuning were used to build a localized large model optimized for imaging report workflows. From November 2023 to March 2024, pilot applications were selected across four modalities — CT, MR, X-ray and ultrasound — with process re-engineering and system integration, followed by trial operation and optimization from April to December 2024. Architecturally, the full DeepSeek general model is deployed at the Beijing Health Cloud Mentougou computing center and used for the knowledge Q&A scenario, while the Shanhai model distilled DeepSeek, pre-trained and fine-tuned on imaging data, powers medical imaging report generation. The report generation scenario integrates with the hospital PACS for voice-based radiology report generation and quality control, while the knowledge Q&A scenario integrates with the OA system and integration platform for knowledge base management, user roles and permission control. Built on RAG (retrieval-augmented generation) technology, the knowledge system integrates hospital regulations, notices, clinical guidelines and other resources into a secure, unified internal knowledge base. From January to April 2025, the hospital promoted the project to other Beijing medical institutions through the Beijing Municipal Health Commission and the Beijing Medical Imaging Quality Control Center.