Le, H. A. | Search bioRxiv

bioRxiv2026

Multicenter self-supervised computational pathology identifies prognostic histomorphological phenotypes in colorectal cancer

H&E whole-slide images capture prognostic information encoded in tumor morphology and the surrounding microenvironment, but these signals remain difficult to extract and interpret at scale. Here, we developed a self-supervised computational pathology framework to predict disease-free survival in colorectal cancer and link model-derived risk to interpretable histomorphology and spatial tumor biology. Using a multicenter developmental cohort spanning colorectal adenomas and invasive colorectal cancer, we trained HPL-PanColon, a self-supervised representation model, to extract tile-level embeddings and identify recurrent histomorphological phenotype clusters across the adenoma-carcinoma spectrum. Compared with general-purpose pathology foundation models, HPL-PanColon yielded representations with reduced institution- and dataset-specific batch effects. We then applied HPL-PanColon to a global survival cohort of 1,024 colorectal cancer patients in a leave-one-institution-out framework, using tile embeddings to train an attention-based survival model and derive the Colon Histomorphology Prognostic Score (CHiPS). CHiPS stratified patients by disease-free survival and provided complementary prognostic information to a UICC TNM-informed clinicopathological model, increasing the c-index from 0.683 to 0.706. Integrating model attention with phenotype assignments traced CHiPS-associated risk to pathologist-recognizable tissue patterns, with high-risk regions enriched for desmoplastic, stromal, and fibroinflammatory morphologies and low-risk regions reflecting tumor-rich epithelial glandular patterns. Spatial transcriptomic analysis further linked high-risk morphologies to fibroblastic, perivascular, myofibroblastic, and immune-reactive tumor microenvironment programs, while low-risk morphologies mapped to epithelial and tumor-enriched regions. These findings establish a scalable framework for interpretable histology-based prognosis and spatial biological discovery in colorectal cancer.

pathology↗

bioRxiv2025

A multi-scale segmentation-free self-supervised AI model to characterize the heterogeneity of the brain tumor microenvironment

Brain tumors affect about 1 million people in the U.S., with aggressive types like glioblastoma having very low survival rates due to complex tumor biology and the protective blood-brain barrier. Current treatments are limited in effectiveness, and our understanding of brain tumor biology remains incomplete. High dimensional multiplexed imaging has enabled us to better understand the tumor microenvironment (TME); however, analyses typically rely on cell segmentation, which is error-prone, may discard useful context outside the cell boundary, and neglects complex tissue-wide features. To address this limitation, we developed a segmentation-free, self-supervised representation learning framework that enables us to train directly on multiplexed images using masked image modeling. We used this approach to analyze 389 imaging mass cytometry images from 185 brain tumor patients. To study tissue-wide features, we first trained our model on 64x64 micron tiles capturing neighborhoods of 10-20 cells, which we termed local tumor microenvironments (LTMEs). To further characterize these LTMEs, we trained our model on 16x16 micron tiles centered on individual cells in our dataset, so that each tile captures a single cell and its surrounding area, which we termed single-cell microenvironments (SCMEs). This multi-scale, self-supervised approach enables a detailed analysis of the heterogeneity within the brain TME, examining single cells in their spatial context. In addition to validating known findings, we identified a novel LTME in GBM patients, composed primarily of tumor cells and a few B and T cells, which strongly correlated with increased survival. By analyzing these B cells with our SCME model, we found they were distinct from other GBM B cells, and higher concentrations of these B cells were linked to improved survival. In conclusion, our study introduces a multi-scale, segmentation-free, self-supervised machine learning model that provides unprecedented insights into brain TMEs, enabling discovery of previously unrecognized cell interactions and spatial features that are predictive of patient survival.

cancer biology↗