Using Unbalanced Regularized Optimal Mass Transport Metrics from DCE-MRI to Classify Radiation Necrosis Versus Tumor Recurrence
Abstract
Purpose
Differentiating radiation necrosis (RN) from tumor recurrence (TR) after radiotherapy remains challenging using conventional MRI. We investigated dynamic contrast-enhanced MRI (DCE-MRI) analysis using unbalanced regularized optimal mass transport (urOMT) to quantify contrast transport properties in lesions and evaluate their ability to distinguish RN from TR.
Methods
DCE-MRI studies from 49 radiation-treated patients with enhanced lesions suspicious for RN/TR were retrospectively analyzed. Lesions were classified as RN (N = 13) or TR (N = 36) based on surgical pathology. The DCE time series were collected with standard imaging protocols at our center, and lesion ROIs were defined by radiologists. The urOMT framework was applied to model spatiotemporal transport properties, producing quantitative transport metrics and visualizations of intralesional heterogeneity. Group differences were assessed using the Wilcoxon rank-sum test. Diagnostic performance was evaluated with ROC analysis; sensitivity and specificity for RN were reported at the Youden-index operating point.
Results
The average contrast density within the ROI was higher in TR than RN (p = 0.0018, AUC = 0.795; sensitivity = 0.846, specificity = 0.722). In contrast, the extreme transport speed was moderately higher in RN than TR, quantified by the mean of the highest 10% contrast speed normalized by density (p = 0.0048, AUC = 0.767 ; sensitivity = 0.846, specificity = 0.611). Contrast clearance flux (efflux) was higher in TR (p = 0.0229, AUC = 0.716).
Conclusion
urOMT analysis of DCE-MRI revealed distinct kinetic and mass-transport signatures for RN and TR: RN showed higher extreme transport speed, which may reflect radiation-induced microvascular injury (e.g., fibrinoid necrosis with luminal narrowing) leading to altered flow dynamics. TR demonstrated higher contrast density and efflux, consistent with increased vascularity and throughput associated with tumor angiogenesis. These urOMT-derived DCE-MRI metrics show potential as quantitative biomarkers for RN vs TR discrimination and require further validation.