Co-clinical CT radiomics pipeline to establish candidate imaging biomarkers for colorectal cancer.

Here, we establish a co-clinical computed tomography (CT) radiomics pipeline for the identification of candidate imaging biomarkers in RAS-mutant metastatic colorectal cancer (mCRC). Orthotopic KRAS-mutant xenograft models (LOVO-Luc2 (N = 52) and SW480-Luc2 (N = 52)) were treated with standard-of-care regimens (FOLFOX, bevacizumab, or combination) and longitudinally imaged by CT (N = 104 tumour scans, N = 156 liver scans collected over 4 timepoints). Radiomic features derived from primary tumour and liver parenchyma were assessed as biomarkers of treatment sensitivity and early metastatic disease. Pre-treatment CT-radiomics identified baseline radiomic correlates of treatment sensitivity in the LOVO-Luc2 model (0.716), with first-order statistical features (Median, 10percentile) significantly associated with therapy outcome.

Texture-based liver radiomic features enabled prediction of metastases earlier than visual CT assessment (AUROC = 0.871). The most predictive features, GLSZM Gray Level Non-Uniformity, GLRLM Run Length Non-Uniformity Normalized, and GLDM Small Dependence Emphasis, were significantly associated with metastatic burden and survival in a clinical CRC cohort (N = 41), indicating species conservation and translational relevance. Collectively, these data demonstrate that preclinical CT radiomics can identify quantitative imaging features associated with treatment sensitivity and early metastatic progression, supporting translational potential.