A research team led by The Hong Kong University of Science and Technology (HKUST) has developed a pioneering artificial intelligence (AI) pathology analysis system that can accurately recognize multiple types of cancer using only a minimal number of samples-without requiring any additional training. This breakthrough significantly enhances the flexibility and efficiency of AI-assisted medical care, marking a major step forward toward the widespread adoption of intelligent pathology.
Each year, nearly 20 million new cancer cases are diagnosed worldwide, with pathological examination playing a pivotal role in clinical diagnosis and treatment decision-making. However, amid a severe global shortage of pathologists, the medical community is increasingly in need of innovative solutions to improve the efficiency of pathological analysis. While AI holds great potential for automating pathological diagnostics, its practical deployment remains constrained by multiple challenges. Conventional AI models typically require the collection and training of tens of thousands of pathology images and datasets to train for each specific cancer type or diagnostic task, resulting in lengthy development cycles and substantial computational and manpower costs. Furthermore, existing foundational pathology models often lack sufficient generalizability, necessitating extensive fine‑tuning when applied across different tumor types in real‑world clinical settings, thereby limiting their scalability and adoption, particularly in resource‑constrained regions.
To address these challenges, a research team led by Prof. LI Xiaomeng, Assistant Professor of the Department of Electronic and Computer Engineering, and Associate Director of Center for Medical Imaging and Analysis at HKUST, in collaboration with Guangdong Provincial People's Hospital and Harvard Medical School, developed a novel pathology analysis system named PRET (Pan‑cancer Recognition without Example Training). The system is the first to introduce the concept of "in-context learning" from natural language processing into pathological image analysis. It allows the model to instantly adapt to new cancer types and perform diagnostic tasks, such as cancer screening, tumor subtyping, and tumor segmentation, during the inference stage by referencing only one to eight annotated tumor slides. Functioning as a "plug-and-play" intelligent diagnostic tool, PRET fundamentally overcomes the need for task-specific fine-tuning in traditional AI models.
The research team conducted comprehensive validation of the PRET system using 23 international benchmark datasets from medical institutions in the Chinese Mainland, the United States, and the Netherlands, covering 18 cancer types and various diagnostic tasks. The results showed that the system outperformed existing methods in 20 tasks, with its Area Under the Curve (AUC)-a measure of diagnostic accuracy-exceeding 97% in 15 of those tasks. Notably, PRET achieved an AUC of 100% in colorectal cancer screening, and an AUC of 99.54% in esophageal squamous cell carcinoma tumor segmentation. In the highly challenging task of lymph node metastasis detection, PRET attained an AUC of approximately 98.71% using only eight slide samples, surpassing the average performance of 11 pathologists, whose AUC averaged approximately 81%. Additionally, PRET demonstrated stable and robust generalizability across different populations and regions with varying levels of medical resources.
Prof. Li Xiaomeng said, "The core value of the PRET system lies in breaking down the traditional barriers of 'massive data and repetitive training,' enabling AI-powered pathology systems to be applied in real clinical settings at lower cost and with greater flexibility. This not only helps alleviate the workload pressure faced by pathologists, but also has the potential to improve access to cancer diagnosis in underserved regions. Through this 'plug-and-play' system, we hope that advanced and precise AI-powered diagnostic services can transcend geographical and resource constraints, thereby promoting global healthcare equity."
Looking ahead, the research team plans to further enhance the system's diagnostic performance and expand its applications to additional clinical tasks, such as genetic mutation prediction and patient prognosis assessment, opening up new directions for the future of AI-driven pathological diagnosis.
The research findings have been published in the international leading journal Nature Cancer.
