Evaluation of MR-Linac Delta-Radiomics Features As Predictors of Response In Hepatocellular Carcinoma
Abstract
Purpose
Radiomics, a field of quantitative characterization of tumor heterogeneity and has shown promise for predicting treatment response in hepatocellular carcinoma (HCC). Delta-radiomics extends this approach by capturing temporal changes in radiomic features (RFs) during therapy. This study evaluated the evolution of delta-RFs derived from daily onboard MRI acquired during MRI-guided radiotherapy for HCC and assessed their association with imaging response.
Methods
A retrospective cohort of 44 HCC lesions from 31 patients treated with MRI-guided radiotherapy was analyzed. Thirty-nine first- and second-order RFs were extracted from gross tumor volumes (GTVs) on daily onboard MRI across five treatment fractions. Delta-RFs were calculated as normalized changes at each fraction relative to the first fraction. Random forest models were used to identify RFs most strongly associated with imaging response, defined by modified RECIST criteria. Predictive performance of the top-ranked features was evaluated using receiver operating characteristic (ROC) analysis with bootstrapping and cross-validation.
Results
GLSZM Small Zones High Gray Level Emphasis and GLSZM Small Zone Emphasis were the most consistently highly ranked delta-RFs across treatment fractions, with one or both appearing among the top two features for each fraction. Random forest models incorporating these features achieved mean area under the ROC curve (AUC) values of 0.755 at fraction 2, 0.436 at fraction 3, 0.711 at fraction 4, and 0.497 at fraction 5.
Conclusion
Delta-RFs derived from daily onboard MRI demonstrate modest predictive capability for HCC treatment response at select time points during MRI-guided radiotherapy. Feature changes observed early in treatment, particularly between the first and second fractions, appear to be the most prognostically informative, suggesting sensitivity to radiotherapy-induced tumor response. These findings agree with previous work that specific treatment time points may yield more meaningful delta-radiomic information and may be helpful in determining patient outcomes with future research.