Prognostic and predictive value of radiomics-based imaging features in patients with colorectal liver metastasis receiving radioembolisation in first-line setting.

Purpose: To evaluate the prognostic and predictive value of radiomics-based imaging markers in colorectal liver metastasis treated with chemotherapy alone or combined with selective internal radiation therapy in the first-line setting. Methods: This was a post-hoc retrospective analysis of the randomized controlled SIRFLOX trial. 491 patients (333 male, median age 63 [range, 28-83] years) with available baseline Computed Tomography (CT) images were included in this analysis. All lesions were segmented automatically in baseline CT with an nnU-net and evaluated against manual segmentation of 80 patients. Quantitative features of tumor segmentations were computed using PyRadiomics. Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to identify relevant prognostic factors, and potential predictive factors were modeled as interaction with the treatment arm. Results: 239 patients had been randomized to FOLFOX alone arm, and 252 patients to the experimental arm. There was no difference in overall survival between treatment arms. A Cox proportional hazards model with LASSO regularization identified 20 prognostic factors. In addition to seven clinical parameters and eight radiomics-based prognostic markers, the LASSO model identified five interaction effects with treatment, highlighting two radiomics features, "shape - Maximum2DiameterSlice" and "glrlm - RunEntropy," as particularly relevant. When patients were categorized into two risk groups based on the model's survival predictions ≥ 50%, patients with high-risk had significantly shorter overall survival than the low-risk group (p < 0.001). Conclusion: Radiomics-based imaging features of liver metastases in pretreatment CT images can identify colorectal cancer patients with poor outcome and potential benefit from combined therapies.