Recently, the inaugural Shanghai Jiao Tong University "Top Ten Vertical Domain Large Models" selection and award ceremony was successfully held at the Zero Bay Technology Transfer Center on the Minhang Campus. This event, hosted by the Shanghai Jiao Tong University Science and Technology Development Research Institute and co-organized by the Zero Bay One-Kilometer Incubator, attracted 37 research teams from various departments across the university to compete. Professor Yu Zhangsheng, founder of Zhiyao Yuanchuang Medical Technology (Shanghai) Co., Ltd., and his team stood out with their Pan-Cancer Pathology Image General Basic Large Model (BEPH), successfully winning first prize, demonstrating the company's cutting-edge innovative strength in the fields of medical artificial intelligence and computational pathology.

This selection process, guided by the "AI+" industrial ecosystem, aimed to identify large-scale models across the university that demonstrate strong technological originality, significant clinical application value, and high industrial applicability. Professor Yu Zhangsheng's team's BEPH model, the first publicly published fundamental model in the field of computational pathology in China, successfully made it onto the university-level AI innovation achievement honor platform thanks to its groundbreaking technology and high-value application potential.

Universal Basic Model for Pan-Cancer Pathology Images (BEPH)

Person in charge: Yu Zhangsheng (Dean of the Institute of Clinical Medicine, Shanghai Jiao Tong University School of Medicine; Tenured Professor, School of Life Sciences and Biotechnology, Shanghai Jiao Tong University)

Team members: Yang Zhaochang, Wei Ting, Liang Ying, Yuan XinGao Ruitian,Xia Yujia, Zhou Jie, Zhang Yue

BEPH, a general-purpose foundational model for pan-cancer pathology images, is the first publicly published foundational model in the vertical field of computational pathology in China (Nature Communications, 2025). This model aims to address the pain points of scarce data annotation, high heterogeneity in cancer morphology, and high computational consumption of existing models in pathological diagnosis. It employs masked image modeling technology and adopts a lightweight approach of "small sample, big intelligence," requiring only approximately 86 million parameters and 11.77 million image patches to achieve deep semantic understanding of 32 cancer pathological features. Its core innovation lies in proposing a local random sampling strategy to improve training efficiency and utilizing masked image modeling technology to capture the deep semantic structure of pathological images, making it more adaptable to high-density pathological features compared to contrastive learning methods. The model demonstrates superior performance in cancer diagnosis (image patch level and whole-slice image level) and survival prediction tasks, and possesses extremely high data utilization efficiency, achieving the performance of similar models with 100% of the data volume using only 50% of the training data. In addition, BEPH has completed the adaptation to the domestic computing power environment and has successfully run on the domestic full-stack software and hardware environment (Ascend 910B + CANN), providing the medical industry with an independent, controllable, cost-effective AI solution with extreme data efficiency.

Professor Yu Zhangsheng is a tenured professor at the School of Life Sciences and Biotechnology, Shanghai Jiao Tong University, and the dean of the Institute of Clinical Medicine, School of Medicine. He is also the founder of Zhiyao Yuanchuang Medical Technology (Shanghai) Co., Ltd. Professor Yu has long been deeply involved in the interdisciplinary field of artificial intelligence, biostatistics, and medical imaging. He has led numerous national and provincial-level research projects and published over a hundred high-level papers in top international journals. He is a leading scholar in the integration of industry, academia, and research in medical AI in China.

This award is a high recognition of Professor Yu Zhangsheng's team's research strength and a significant milestone for Zhiyao Yuanchuang in the development and industrialization of large-scale medical models. In the future, Zhiyao Yuanchuang will continue to leverage the cutting-edge research achievements of Professor Yu Zhangsheng's team as its core engine, driving the BEPH large-scale model and its series of medical AI technologies from the laboratory to clinical practice and industry. Focusing on core areas such as precision medicine, intelligent pathology, and new drug development, Zhiyao Yuanchuang will provide the industry with more efficient and reliable artificial intelligence solutions, empowering the high-quality development of the healthcare industry through technological innovation!

Article link:

Yang, Z., Wei, T., Liang, Y. et al. A foundation model for generalizable cancer diagnosis and survival prediction from histopathological images. Nat Commun 16, 2366 (2025).https://doi.org/10.1038/s41467-025-57587-y