Colorectal cancer frequently spreads to the liver, and assessing the response of colorectal liver metastases (CRLM) to chemotherapy is a major determinant in treatment planning. Accurate non-invasive assessment of tumour regression grade (TRG) is vital for guiding surgical and systemic therapy decisions. While radiologic evaluations like RECIST often overestimate therapeutic response, radiomics—an emerging field of quantitative imaging—offers a promising tool for enhancing prediction accuracy. A multicentre retrospective study in Italy explored how combining clinical parameters with radiomic features extracted from post-chemotherapy gadoxetic acid–enhanced MRI can improve the prediction of pathologic response in CRLM.
Radiomic Integration Improves Predictive Accuracy
The study analysed 162 patients who underwent liver resection for CRLM following oxaliplatin- or irinotecan-based chemotherapy between 2018 and 2021. Each participant underwent a preoperative MRI scan within 60 days of surgery, and the largest metastasis per patient was segmented on portal venous phase (PVP) and hepatobiliary phase (HBP) sequences. Radiomic features were extracted from both the tumour volume (Tumour-VOI) and a 5-mm peritumoural margin (Margin-VOI). When combined with clinical variables such as CEA levels, metastasis synchronicity and treatment details, radiomic data significantly improved model performance.
Must Read: Comparing MRI Protocols for Colorectal Liver Metastases Detection
The combined clinical-radiomic model achieved a validation accuracy of 0.773, with a sensitivity of 0.724 and a specificity of 0.812. The ROC-AUC of 0.860 was significantly higher than the clinical model alone. Radiomic features from the Tumour-VOI in the PVP and Margin-VOI in the HBP were especially influential. These findings suggest that MRI-based radiomics can detect subtle tissue heterogeneity associated with pathological response, thus refining therapeutic decision-making beyond what conventional radiology offers.
Tissue and Imaging Phase-Specific Insights
The added value of radiomics lies in its ability to quantify tumour heterogeneity, necrosis and peritumoural changes that are not readily visible through standard imaging. In the PVP sequence, features like intensity range positively correlated with TRG1–3, likely reflecting intratumoural areas of fibrosis and necrosis—hallmarks of treatment response. Conversely, in the HBP sequence, a higher intensity range negatively correlated with TRG, potentially indicating aberrant contrast retention due to expanded extracellular space or altered transporter expression.
Shape-related metrics and texture features such as mesh volume and GLSZM-based indices also contributed significantly. Moreover, the peritumoural margin proved particularly informative in the HBP. Features like energy in this region were associated with better responses, possibly due to correlations with immune infiltration or desmoplastic reactions. These subtle but quantifiable patterns suggest radiomics can provide biologically meaningful insights into tumour behaviour and treatment effects.
Clinical Implications and Future Integration
The implications of the study extend to more accurate selection of patients for surgery and systemic therapy. By identifying non-responders who may otherwise appear to benefit from treatment on conventional imaging, clinicians can better stratify risk and modify treatment plans. Radiomics could also identify early reactivations of disease not captured by RECIST criteria. This is particularly relevant as nearly 45% of patients with radiologically defined partial responses were found to have no histological regression.
Despite the promising results, several limitations must be addressed before clinical implementation. The retrospective nature of the study and the exclusion of T2-weighted and DWI sequences due to inter-centre variability limit generalisability. Manual segmentation, though reproducible in the study, remains labour-intensive and subject to variability. Additionally, the absence of external validation and a user-friendly scoring tool for real-world application necessitate further research. Prospective trials with standardised imaging protocols and automated analysis pipelines will be essential to transition this technique from research to routine care.
MRI-based radiomics, when combined with clinical variables, enhances the prediction of pathologic response in patients with CRLM undergoing systemic therapy. This integrated approach can significantly improve treatment planning, offering a more personalised strategy for managing liver metastases from colorectal cancer. The study supported the growing role of advanced imaging analytics in precision oncology, highlighting the need for prospective validation and workflow integration to ensure broader clinical adoption.
Source: European Journal of Radiology
Image Credit: iStock
