Radiological and pathological integration predicts the efficacy of bevacizumab with chemotherapy in colorectal cancer liver metastases

Abstract While bevacizumab plus chemotherapy is considered the standard therapy for unresectable colorectal cancer liver metastases, a substantial proportion of patients exhibit inadequate responses. This underscores the pressing need for reliable predictive biomarkers. Recent advances in radiomics and pathomics offer robust frameworks for developing sufficiently accurate signatures. However, the development of predictive models for bevacizumab efficacy and their biological interpretations remain to be explored. Here, we developed a machine learning-based integrative model combining advanced abdominal CT imaging with digitized H&E-stained tissue slides to predict bevacizumab efficacy. Validation across multi-center internal and external cohorts confirmed the model's clinical robustness. The radiomic model identified peritumoral features as providing the most significant predictive contribution among various regions. Model interpretation using paired transcriptome data before and after treatment revealed that fibroblasts adjacent to the tumor may mediate therapeutic resistance by promoting an inflammatory microenvironment. Coupling dimensionality reduction of deep learning-extracted histopathological features with single-cell RNA sequencing identified fibroblasts as the critical mediators of treatment resistance. By integrating radiomics and pathomics, we developed a dual-omics model to enhance predictive performance and enable efficacy stratification. Collectively, our study established an accurate predictive model that integrated imaging analysis and histopathological profiling and elucidated the tumor microenvironment heterogeneity related to therapeutic outcomes. By integrating radiological and pathological signatures with fibroblasts and peritumoral angiogenesis, this work emphasizes the pivotal role of stromal factors in driving drug resistance, providing practical insights for the accurate stratification and early intervention strategies in colorectal cancer liver metastases.