Recently, the DHC Rare Disease Large Model Technology Application Seminar was successfully held at the Beijing headquarters of DHC. The seminar brought together numerous experts and scholars from top medical institutions across the country, including Professor Qiu Zhengqing and Teacher Zhang Zhenjie from Peking Union Medical College Hospital, Professor Meng Yan and Teacher Zhang Qi from the General Hospital of the People's Liberation Army, Professor Yang Yanling and Professor Dong Hui from Peking University First Hospital, Professor Zhang Ningping from Zhongshan Hospital affiliated with Fudan University, Professor Lu Yulan from the Pediatric Hospital affiliated with Fudan University, Professor Zhang Zhiyong from the Children's Hospital affiliated with Chongqing Medical University, Deputy Director Lin Hao from the Affiliated Hospital of Guangdong Medical University, Deputy Director Miao Huilai and Deputy Director Zhou Haihong from the Second Affiliated Hospital of Guangdong Medical University, Deputy Director Lai Tianwen from the First Dongguan Hospital affiliated with Guangdong Medical University, Secretary Zhong Bomao from the Dongguan Children's Hospital affiliated with Guangdong Medical University, President Chen Yongsong from the First Affiliated Hospital of Shantou University Medical College, and President Chen Riling from the Shunde Women and Children's Hospital affiliated with Guangdong Medical University. These renowned experts and scholars in their respective fields participated in the meeting both online and offline.

At the seminar, Dr. Xu Juan, Vice President and Dean of the Artificial Intelligence Research Institute of DHC, provided a detailed introduction to the DHC Rare Disease Large Model. As a leading enterprise in medical big data and artificial intelligence, DHC has been deeply involved in the field of rare diseases for many years, assisting Peking Union Medical College Hospital in establishing China's first national-level rare disease registration platform - the National Rare Disease Registration System (NRDRS). By the end of 2023, NRDRS had established 193 rare disease research cohorts, covering 176 types/categories of rare diseases, with over 70,000 cases from 104 cooperative units across 29 provinces, cities, and autonomous regions nationwide, laying a solid foundation for the development of the rare disease large model.

The DHC Rare Disease Large Model is the first multimodal solution specialized in rare diseases in China, covering applications such as intelligent Q&A, research assistant, doctor assistant, and decision support. Dr. Xu Juan emphasized that based on the large model foundation of DHC, the advanced rare disease specialty Q&A large model was trained. The large model has learned from more than 130 textbooks, guidelines, consensuses, and over 10,000 literatures in the field of rare diseases, constructing a specialty knowledge base of over 700,000 entries, with more than 30,000 image-text pairs, fully covering all 207 types/categories of rare diseases listed by the National Health Commission, State Drug Administration, and Traditional Chinese Medicine Administration.

Professional domain deep cultivation and data accumulation: The deep involvement of the DHC Rare Disease Large Model in professional domains is reflected in its extensive integration of medical professional corpus, including medical guidelines, consensuses, textbooks, literature, and real case information. This depth of data accumulation provides the model with a rich background of professional knowledge, making it more professional and precise in handling rare disease-related issues.

Multimodal data processing and analysis capabilities: Another significant advantage of the DHC Rare Disease Large Model is its ability to process multimodal data. It can not only handle text data but also integrate imaging, pathology, and genetic data, providing comprehensive disease analysis, diagnostic support, treatment recommendation, and patient management advice. This cross-modal data fusion enables the model to offer richer and deeper insights when assisting doctors in complex case analysis, helping doctors make precise diagnoses and differential diagnoses, and improving diagnostic efficiency.

Research assistant and knowledge base construction: The application of the DHC Rare Disease Large Model in the research field also demonstrates its strong capabilities. It can quickly search and summarize a large amount of literature, mine research hotspots, assist researchers in building research ideas, write research papers and draw diagrams, enhancing the efficiency and quality of research work. Additionally, the model supports the construction of unique knowledge bases, phenotype databases, and maps for rare diseases, providing clinical experts with quick access to professional diagnostic and treatment advice.

Following this, Dr. Zhou Wenzhong from the Artificial Intelligence Research Institute of DHC presented an exciting demonstration on the construction process of the rare disease Q&A large model.He elaborately introduced the technical architecture and evaluation process of the model, including the ability assessment in three stages: popular science knowledge, professional scenarios, and diagnostic case analysis, as well as the scoring standards in five dimensions: relevance, accuracy, conciseness, logic, and traceability.

In comparison with other domestic and international large models, the DHC Large Model stands out in terms of accuracy and logic, especially in comprehensiveness regarding genetic mechanisms, pathological mechanisms, diagnosis, and treatment. Furthermore, the DHC Model also shows higher accuracy and conciseness in professional scenario Q&A and diagnostic case analysis. Dr. Zhou also pointed out that the DHC Model could be applied not only in Q&A systems but also play a significant role in multiple fields such as medical literature retrieval, knowledge acquisition, assisted diagnosis and treatment, and patient consultation management, demonstrating its wide application potential and practical value in the medical field.

During the expert review session, President Chen Yongsong, Professor Zhang Zhiyong, Deputy Director Lin Hao, and Professor Dong Hui each gave high praise to the technology of the DHC Rare Disease Large Model from their perspectives and offered valuable comments and suggestions. President Chen Yongsong emphasized the importance and urgency of rare disease research, while Professor Zhang Zhiyong appreciated the model's application in literature retrieval and knowledge management from the perspective of a research assistant. Deputy Director Lin Hao and Professor Dong Hui discussed the practicality and educational value of the model from the perspectives of clinical application and patient needs.

The DHC Rare Disease Large Model adopts an advanced technical architecture, including a large language model based on Transformer and multimodal learning technology. The application of these technologies makes the model more efficient and accurate in handling complex medical issues. Additionally, the model possesses the ability to continuously learn online, continuously absorbing new medical knowledge and data to maintain its leading position in the medical field. Experts present unanimously agreed that the DHC Rare Disease Large Model, with its strong professionalism, comprehensive data processing, advanced technical architecture, powerful clinical support, and outstanding research capabilities, provides strong AI support for the diagnosis and treatment of medical fields, especially rare diseases. These advantages not only drive the development of medical technology but also offer new possibilities for improving the quality and efficiency of medical services.