The Hong Kong University of Science and Technology recently announced the development of four major language models for healthcare.
WHAT IS IT ABOUT
LLM-based tools developed at HKUST SuperPOD, the university’s AI supercomputing facility, include:
-
MOTHER: AI model for identifying breast cancer pathology in MRI images.
-
mSTAR: Pathology assistant tool that directly models entire slide images.
-
MedMr: A multimodal chatbot language model.
-
WIRE: An explainable AI model that provides visual and textual explanations for AI-generated analytics.
In a statement to the HKUST research team shared further details and conclusions from the development of these models.
Their breast cancer AI model achieved 87% diagnostic accuracy in multi-center testing. MOME can also predict patient response to neoadjuvant chemotherapy.
According to the team, mSTAR solves “a wide range of clinical pathology problems, including seven types of clinical diagnostic and prognostic applications such as tumor subtype and stage, metastasis detection, molecular prediction, survival analysis and report generation.”
The MedMR medical chatbot, which answers questions, generates medical reports and makes preliminary diagnoses based on medical images, achieved 93% accuracy in identifying cancer and non-cancer using the publicly available PCam200 dataset from Patch Camelyon.
Moreover, they reported that XAIM achieved 98.67% accuracy in diagnosing skin lesions on the PH² dataset from Portugal.
The research team is now preparing to test these models in various hospitals in the city. “We are in talks with several hospitals in Hong Kong about potential trials and implementations.”
“Before AI models are implemented in clinical practice, we will also need to work with hospitals to conduct large-scale, multi-center validation of the models to ensure their generalizability and reliability,” they added.
A BIGGER TREND
Earlier this year, the Hong Kong Center for Artificial Intelligence and Robotics, a research center under one of China’s national research institutes, the Chinese Academy of Sciences, also introduced a doctor chatbot called CARE. The chatbot, built on Meta’s Llama 2 LLM, is being tested in seven anonymous hospitals in Beijing.
Other Asian healthcare systems have been working on similar generative AI projects. In October Singapore’s Ministry of Health has announced a novel investment to support a national project to implement genAI in the public health system by the end of 2025. Singapore-based Docquity is also helping community health centers in West Java, Indonesia to implement the genAI virtual assistant TehaAIsupporting healthcare workers in diagnosing tuberculosis, dwarfism and hypertension.