From Single-Cancer to Pan-cancer Prognosis: A Multi-modal Deep Learning Framework for Survival Analysis with Robust Generalization Capability

Binyu Zhang; Shichao Li; Junpeng Jian; Xiaoyu Ren; Ziqi Zhao; Limei Guo; Fei Su; Zhu Meng; Zhicheng Zhao

doi:10.1016/j.ajpath.2025.06.006

From Single-Cancer to Pan-cancer Prognosis: A Multi-modal Deep Learning Framework for Survival Analysis with Robust Generalization Capability

Am J Pathol. 2025 Jul 10:S0002-9440(25)00238-X. doi: 10.1016/j.ajpath.2025.06.006. Online ahead of print.

Authors

Binyu Zhang¹, Shichao Li¹, Junpeng Jian¹, Xiaoyu Ren¹, Ziqi Zhao², Limei Guo², Fei Su³, Zhu Meng⁴, Zhicheng Zhao⁵

Affiliations

¹ School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China.
² Department of Pathology, School of Basic Medical Sciences, Third Hospital, Peking University, Beijing, China.
³ School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China; Beijing Key Laboratory of Network System and Network Culture, Beijing, China.
⁴ School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China; Beijing Key Laboratory of Network System and Network Culture, Beijing, China. Electronic address: bamboo@bupt.edu.cn.
⁵ School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China; Beijing Key Laboratory of Network System and Network Culture, Beijing, China. Electronic address: zhaozc@bupt.edu.cn.

PMID: 40651540
DOI: 10.1016/j.ajpath.2025.06.006

Abstract

Accurate prognosis represents a critical component in oncology research, enabling personalized treatment planning, and optimized healthcare resource utilization. While existing prognostic models demonstrate promising performance on restricted datasets, they remain constrained by two limitations: modality-specific architectural designs and cancer type-specific training paradigms that hinder cross-domain generalization. To address these challenges, the Unified Multi-modal Pan-cancer Survival Network (UMPSNet) is introduced, which integrates histopathology images, genomic expression profiles, and four metadata categories through structured text templates. UMPSNet employs the optimal transport (OT)-based attention for multi-modal feature alignment and a guided mixture of experts (GMoE) mechanism to address cancer-type distribution shifts. Comprehensive evaluation across 3,523 whole slide images (WSIs) (n=2,831) spanning five TCGA cohorts demonstrated superior predictive performance (mean C-index=0.725), surpassing meticulously designed single-cancer models. Notably, in zero-shot transfer evaluation involving 392 pancreatic adenocarcinoma WSIs (n=66) from Peking University Third Hospital, UMPSNet achieved a C-index of 0.652 without parameter fine-tuning, demonstrating generalization capacity for previously unseen malignancies. Additionally, UMPSNet identified prognostic gene signatures that consistently overlapped with clinically detected mutations (n=92) while revealing novel gene candidates, validating its clinical relevance and providing complementary insights for precision oncology. The UMPSNet framework establishes a new paradigm for multi-modal survival analysis by overcoming data heterogeneity and domain shift challenges, thereby providing a clinically adaptable tool for pan-cancer prognostic prediction.

Keywords: Deep learning; Multi-modal integration; Pan-cancer prognosis; Zero-shot generalization.