Histopathology-Based Spatial Profiling of Immune and Molecular Features Predicts Cancer Risk in Barrett’s Esophagus Patients

Abstract

Barrett's esophagus (BE) patients require improved cancer risk stratification to identify high-risk individuals while reducing over-treatment in low-risk populations. BE evolution towards adenocarcinoma is likely driven by a combination of genomic and microenvironmental factors, suggesting the potential of predictive models that integrate both. Here, we developed BEACON (Barrett’s Esophagus DNA content Abnormality and immune ecology for Cancer Outcome), a spatially aware framework integrating deep learning-based prediction of DNA content and immune architecture from routine histopathology. First, DACOR, a multi-instance learning model that predicts DNA content from histopathology, was trained using 777 BE biopsies with flow cytometry-based DNA content data scanned at two institutions, and it demonstrated strong performance across both validation and test datasets for DNA content abnormality prediction. Patients classified as DNA content abnormal by DACOR demonstrated increased cancer progression. Complementary models for cell classification and tissue segmentation enabled spatial immune ecology metric computations. Abnormal regions exhibited increased lymphoplasmacytic inflammation versus normal regions. Among patients with DNA content abnormality, cancer progressors exhibited increased plasma cell clustering adjacent to abnormal epithelium compared to non-progressors. Finally, a logistic regression model was developed that integrated molecular, immune ecological, and epithelial morphology features for cancer risk stratification. The integrated risk classification model stratified DNA content abnormal patients into high- and low-risk groups that corresponded with progression and patient survival. Overall, BEACON spatially integrates molecular abnormality with immune ecology to stratify BE patients by cancer progression risk using routine pathology images.