Hong Kong-based Centre for Artificial Intelligence and Robotics (CAIR) under the Chinese Academy of Sciences (CAS), one of China's national research institutes, has developed a large language model-based chatbot to assist doctors in making medical diagnoses and treatments.

HOW IT WORKS

The chatbot, CARES Copilot 1.0, is built based on Meta's Llama 2 LLM. Trained using 100 GPUs from Huawei and Nvidia, it can process various data, including images, text, voice, video, MRI, CT scans, and ultrasound. 

Fed with millions of records, including teaching materials, expert guidelines, and medical literature, CARES can retrieve information "within a few seconds," with up to 95% accuracy, CAIR claims. CARES, equipped with a 100K context window, can also pull out information from over 3,000 pages of surgical materials.

Based on internal testing, it can support functions such as surgical phase identification, segmentation of instruments and anatomical structures, instrument detection and counting, and generation of MRI high-resolution images. 

The assistant AI tool, which can be integrated with medical devices, is now undergoing continuous optimisation in seven hospitals in Beijing. 

THE LARGER TREND

China is playing catch up with global innovators in developing AI technologies by funding local initiatives. CAIR's latest AI development is one of those projects co-funded by the Hong Kong government's InnoHK research programme. In August last year, the Chinese government approved the rollout of the first homegrown genAI-enabled service, the TaiChu model, also by CAS.

ON THE RECORD

"[LLMs], tailored for neuroscience, can surpass traditional teaching needs, enabling applications in clinical settings, operating rooms, and research institutes in conjunction with surgery, image guidance, and robotics. These models directly support frontline medical staff in emergencies by supervising, issuing warnings, preventing risky procedures, and pushing the boundaries of neuroscience. The combination of LLMs and surgical navigation can offer real-time anatomical positioning information to surgeons, enhancing surgical safety," Ming Feng, professor and chief physician at the Department of Neurosurgery, Peking Union Medical College Hospital, commented on the release of CARES.