Cancer diagnosis often takes a long time, but the shortage of doctors in Hong Kong and the mainland affects efficiency. In light of this, HKUST has developed an AI pathology system called SmartPath, which provides one-stop cancer diagnosis support, integrates patient data, analyzes complex pathological images, and empowers doctors in cancer diagnosis, prevention, and improving diagnostic efficiency.
The model has been in use in the radiology department for six months. At Nanfang Hospital in the mainland, preparations are underway to conduct further clinical trials in collaboration with institutions over the next six months to a year, in preparation for broader application.
According to Professor Chen Hao, Assistant Professor in the Department of Computer Science and Engineering and Director of Collaborative Center for Medical and Engineering Innovation at HKUST, conventional cancer diagnosis relies on imaging examinations and histopathological examinations. The latter is considered the "gold standard" for cancer diagnosis, with specific operations including the diagnosis of pathological sections.
Can assist in completing over 100 clinical tasks
Based on clinical experience, a senior doctor takes about 3 to 5 minutes to complete the diagnosis of a single slice, while a junior doctor might need over 10 minutes to complete the same task. Chen stated that user study data previously conducted showed that with the assistance of SmartPath, the diagnostic efficiency of pathologists can be improved by over 30%. Simultaneously, the system can effectively optimize diagnostic quality, increasing diagnostic consistency among different doctors by over 10%, providing strong support for reducing clinical diagnostic discrepancies and ensuring standardized diagnosis and treatment.
Regarding the SmartPath system itself, Chen further introduced that it is built upon the world's largest and most diverse pathological dataset and can assist healthcare workers in completing over 100 clinical tasks, including cancer grading and typing.
Over the past few years, the team has developed two major pathological foundation models: the first is the visual technology model GPFM, and the second is the world's first multimodal foundation model, mSTAR. By enhancing multimodal information to compensate for the limitations of single visual information, it achieves comprehensively best performance in multiple tasks and can accurately diagnose high-incidence cancers in Hong Kong, such as lung cancer and breast cancer. mSTAR is the world's first model to achieve multimodal information enhancement and has attained the best comprehensive performance across multiple tasks.
Clinical validation targeting four major cancer types
Currently, SmartPath has undergone prospective clinical validation for four major cancer types: lung cancer, gastric cancer, breast cancer, and colorectal cancer. The average diagnostic accuracy for colorectal cancer exceeds 95%, reaches 93% for gastric cancer, and also exceeds 93% for lung cancer.
Prof. Liang Li, Director of the Department of Pathology, Nanfang Hospital, School of Basic Medical Sciences, Southern Medical University, stated, "AI models can play a key role in clinical decision-making for tumors like lung cancer, potentially avoiding unnecessary surgeries and ineffective post-operative treatments."
When asked about the project's next development plans, Chen said that SmartPath has already completed prospective clinical trials for the four major cancer types at Nanfang Hospital in the mainland. The next step, within the next six months to a year, involves collaboration with other institutions to prepare for promotion. Simultaneously, the system will continue to be optimized in practical use, employing a "spot-the-difference" fine-tuning approach: for example, after inputting 1000 cases, if the model shows deviations in diagnosing invasive cancer, specifically supplement with 1000 to 2000 invasive cancer cases, allowing the model to continuously learn and correct itself.
(Source: Ta Kung Pao; Journalist: Qiu Ziyin; English Editor: Darius)
Related News:
Deepline | Watch your step! HK boosts mountain safety as rescue calls rise ahead of hiking season
Deepline | Digital immune system: How HK banks are fighting AI with AI
Comment