CT radiomics for survival risk stratification in resectable colorectal liver metastases: a multi-centre study.
Authors
Affiliations (8)
Affiliations (8)
- Nuffield Department of Surgical Sciences, University of Oxford, John Radcliffe Hospital, Oxford, OX3 9DU, UK. [email protected].
- Department of Hepato-Pancreato-Biliary (HPB) Surgery, Oxford University Hospitals NHS Foundation Trust, Oxford, OX3 9DU, UK. [email protected].
- Human Liver Research Facility, University of Liverpool, Liverpool, UK.
- Department of Hepatobiliary Surgery, Liverpool University Hospitals NHS Foundation Trust, Liverpool, UK.
- Department of Pharmacology, Institute of Systems, Molecular and Integrative Biology, University of Liverpool, Liverpool, UK.
- Nuffield Department of Surgical Sciences, University of Oxford, John Radcliffe Hospital, Oxford, OX3 9DU, UK.
- Department of Hepato-Pancreato-Biliary (HPB) Surgery, Oxford University Hospitals NHS Foundation Trust, Oxford, OX3 9DU, UK.
- Department of Molecular and Clinical Cancer Medicine, Institute of Systems, Molecular and Integrative Biology, University of Liverpool, Liverpool, UK.
Abstract
Outcomes following resection of colorectal liver metastases (CRLM) remain heterogeneous, and there is a need for preoperative tools that support survival risk stratification and clinical decision making. CT based radiomics offers a non invasive approach to quantify tumour and liver characteristics using routinely acquired imaging. Here we report the largest multicentre study evaluating the prognostic utility of preoperative CT radiomics in patients undergoing resection of colorectal liver metastases. Patients undergoing curative-intent CRLM resection were retrospectively identified from two UK centres. Radiomic features were extracted from portal-venous CT scans of metastases and background liver. Feature selection techniques were applied. Machine learning models incorporating clinical and/or radiomic variables were trained to predict 3-year mortality. A Radiomic Risk Score (RRS) was derived from six radiomic features and evaluated using Cox regression and Kaplan–Meier analysis in an independent validation cohort. A total of 959 metastases from 399 patients were analysed. Six radiomic features (one morphological, one first-order, and four textural) were retained. Radiomics alone demonstrated moderate discrimination for 3-year mortality (AUC 0.77, 95% CI 0.68–0.85), comparable to clinical models (AUC 0.82, 95% CI 0.75–0.86). Combined clinical–radiomic models demonstrated modest but consistent improvement in performance (random forest AUC 0.83, 95% CI 0.78–0.88). At both 2-year and 3-year timepoints, formal comparison using the DeLong test demonstrated no statistically significant difference between radiomics and clinical models. The RRS stratified patients into high- and low-risk groups with significantly different survival in both the training cohort (HR 2.71, 95% CI 1.86–3.96, <i>p</i> < 0.001; median OS 32 vs. 53 months) and validation cohort (HR 2.01, 95% CI 1.14–3.54, <i>p</i> < 0.001; median OS 24 vs. 38 months). Multivariable analysis confirmed the RRS as an independent prognostic factor (HR 2.91, 95% CI 1.93–4.39, <i>p</i> < 0.001). Preoperative CT radiomics provides complementary prognostic information in patients undergoing resection of colorectal liver metastases. While radiomics alone does not outperform clinical variables for fixed-horizon prediction, it enables robust survival risk stratification and improves combined model performance. Radiomics could be considered as an adjunct to, rather than a replacement for, established clinical assessment in CRLM. The online version contains supplementary material available at 10.1038/s41598-026-48659-0.