China Medical University Hospital leverages MedLM Gen AI to aid physicians

Google Cloud and CMUH celebrate their collaboration

In collaboration with Google Cloud, China Medical University Hospital (CMUH) in Taiwan has developed an “AI-assisted physician” system using Generative AI technology including MedLM, a large language model built on Med-PaLM 2, that aims to support healthcare professionals in disease diagnosis, treatment planning, patient education, and medical research.

The collaboration will introduce assistive tools including “Customized Cancer Treatment Guidelines” and “Cancer Therapy Q&As” that aim to reduce the time spent searching for information and enable the quick capture of accurate data. According to CMUH, this can aid professionals in developing cancer treatment plans, offering personalised treatment-related information to patients, and providing responses to patient health education inquiries.

Prof. Oscar K. Lee, Vice Superintendent of CMUH explained that among global Gen AI LLMs, Med-PaLM 2 stands out in model performance, scoring 85% on MedQA-USMLE (US Medical Licensing Examination), and reaching a level of excellence comparable to professional physicians. By training the model to analyse treatment guidelines and care cultures from various regions, MedLM is expected to become better at addressing the medical and healthcare needs of local populations.

Existing collaborative applications between CMUH and Google Cloud include “Customized Cancer Treatment Guidelines” based on National Comprehensive Cancer Network (NCCN) Guidelines and integrated with MedLM. Clinicians may quickly capture accurate information based on cancer staging, treatment methods, and risk assessment.

CMUH is furthermore leveraging Google’s unique AI accelerators, Tensor Processing Units (TPUs) to assist in calculations related to protein folding and the development of new drugs. Preliminary tests at CMUH have shown that using these resources can reduce the computation time for related programmes by more than tenfold.

Matt avatar

Leave a Reply

Your email address will not be published. Required fields are marked *