Tissue‐Agnostic Cellular Morphometric Biomarkers for Risk‐Adapted Management Across Gastrointestinal Precancerous Lesions and Cancers

Abstract

ABSTRACT While precision oncology increasingly adopts tissue‐agnostic paradigms, current strategies remain heavily reliant on molecular alterations, with limited relevance to early‐stage cancers and precancerous lesion management. Here we present an unsupervised and interpretable artificial intelligence framework that defines tissue‐agnostic cellular morphometric biomarkers (CMBs) capturing conserved tumor microenvironment (TME) architectures associated with cancer progression across gastrointestinal (GI) organs. Discovered from colorectal cancer whole‐slide images and validated in gastric and esophageal malignancies within a multi‐center cohort of 2,602 patients, a 13‐CMB signature demonstrates robust cross‐GI tissue transferability and prognostic impact. Importantly, beyond prognostic value in pan‐GI cancers, the 13‐CMB signature enables risk stratification of precancerous lesions and early‐stage cancers, addressing a key unmet need in clinical decision‐making where molecular profiling is often impractical or insufficient. The CMB‐based risk scores support risk‐adapted management, including individualized surveillance intervals and tailored intervention strategies. Integrative analyses using bulk and single‐cell RNA sequencing and immunohistochemistry profiling show that CMBs correspond to biologically interpretable morphological states of the TME, including immune‐excluded and stromal‐dominant architectures. Together, this study establishes a biologically transparent, tissue‐agnostic CMB framework that links conserved TME organization to clinically actionable risk assessment, providing a scalable approach for early cancer management and precision oncology.