HONG KONG, Oct. 31, 2024 /PRNewswire/ -- In a groundbreaking initiative, The Hong Kong University of Science and Technology (HKUST) has unveiled four AI-driven models poised to transform the medical and healthcare fields. These AI models can assist both general and specialist medical practitioners in diagnosing and prognosing up to 30 types of cancers and diseases, with some achieving accuracy comparable to that of medical experts with five years of experience or more.
Supported by the University's new AI supercomputing facility, which offers robust computing power, these large AI models surpass many existing systems due to their foundation on extensive data sets and novel machine training strategies. Prof. CHEN Hao, Assistant Professor in the Department of Computer Science and Engineering and project lead, said that one model processed over 160 million images across 32 cancer types for pathological diagnostic tasks.
The four models include:
- MOME (Breast Cancer Diagnosis)- This is the first adaptation of a large foundation model for 3D mpMRI fusion. Targeting breast cancer, one of the most prevalent cancers among females in Hong Kong, MOME distinguishes malignant from benign breast lumps using MRI scans, achieving accuracy levels comparable to radiologists with 5 or more years of experience. It provides an alternative to the more intrusive diagnostic form of biopsies. Additionally, it predicts patients' responses to neoadjuvant chemotherapy, enabling non-invasive and personalized management of breast cancer.
- mSTAR (Pathology Assistant Tool) - Pathological examination is the gold standard in cancer diagnosis, but generating pathology reports can be time-consuming and prone to errors. mSTAR, recognized as one of the world's leading foundation models, aims to streamline this process. Unlike existing ones that analyze slides by splitting them into individual patches, mSTAR stands out by directly modelling whole slide images and augmenting with multimodal knowledge, further enhancing accuracy. It assists pathologists in performing up to 40 diagnostic and prognostic tasks, significantly reducing examination time and improving diagnostic consistency.
- MedDr (Generalist) - This versatile AI medical generalist functions like a "medical GPT". The multimodal language model can answer questions, generate medical reports, and provide initial diagnoses based on medical images. As one of the largest open-source generalist foundation models tailored for medicine, MedDr helps physicians make faster and more accurate diagnoses. A recent study by Shanghai AI Lab recognized it as one of the best existing generalist models.
- XAIM (Explainable AI) - This novel framework enhances healthcare professionals' confidence in adopting AI by explaining how medical AI models reach their decisions. While many AI models offer high accuracy, they often lack transparency, leading to skepticism. XAIM addresses this by providing visual and textual explanations of the models' analyses.
Prof. Chen, also the Director of the Collaboration Center for Medical and Engineering Innovation, jointly established between HKUST and Union Hospital, Tongji Medical College of Huazhong University of Science and Technology, stated, "We hope our AI tools will serve as valuable assistants for doctors, improving diagnostics, personalizing treatment plans, and streamlining different tasks. Building on our success, we are developing a wide range of AI applications targeting various clinical tasks, particularly precision oncology and computer-assisted intervention, while continuously enhancing our models with more data and modalities. By collaborating closely with clinical partners, we aim to drive improvements in patient care over the long term."
As a leader in AI research, HKUST has made significant contributions to medicine, including an AI model predicting the prognosis of brain tumor patients, a highly accurate blood test for early detection of Alzheimer's disease and mild cognitive impairment, and a microscope aiding near real-time judgments in lung cancer surgery, sparing patients' unnecessary procedures.
Video:
Download photos here: