Radiomic Artificial Intelligence Models Predicting the Response of Colorectal Cancer Liver Metastases to Chemotherapy-A Systematic Review.
Authors
Affiliations (3)
Affiliations (3)
- Department of Surgery, Western Precinct, University of Melbourne, Melbourne, Australia.
- Western Health Library Service, Western Health, Melbourne, Australia.
- Department of Colorectal Surgery, Western Health, Melbourne, Australia.
Abstract
Colorectal cancer liver metastasis (CRCLM) profoundly impacts the overall survival rates of affected individuals. Although chemotherapy remains a cornerstone of treatment, the variability in tumour response presents a considerable challenge to effective clinical management. Radiomics-based artificial intelligence (AI) modelling, utilising pre-treatment imaging, has been proposed as a tool to predict treatment response and inform therapeutic decision-making. A systematic search using Clarivate (Web of Science) and Ovid (Embase and MEDLINE) from inception to April 8, 2025, was performed in accordance with Preferred Reporting Items for Reviews and Meta-Analysis (PRISMA) guidelines. Studies evaluating the use of AI radiomics models to predict CRCLM response to chemotherapy treatment, alone or in combination with targeted therapies were identified. Full-text review and data extraction were performed independently by two reviewers. Risk of bias was assessed using the Checklist for Diagnostic Test Accuracy Studies. Thirteen studies met inclusion criteria. All studies described the development and validation of AI models with no implementation studies. The AI models developed demonstrated a varied ability to predict CRCLM response to chemotherapy. Predictive performance was classified as good in six studies and excellent in three. Model development used differing combinations of imaging modalities (CT or MRI), lesion inclusion criteria, dimensionality (2D vs. 3D), and use of clinical features. Most studies did not validate model performance in an external dataset. Radiomics-based AI models show potential in predicting chemotherapy response in CRCLM. However, limited external validation and methodological variability currently restrict clinical applicability. Standardisation and prospective validation are required for clinical translation.