ESTRO 2025 - Abstract Book

S714

Clinical - CNS

ESTRO 2025

Genova, Italy. 4 Radiation Oncology Department, Pitié-Salpêtrière Hospital, Sorbonne University, AP-HP, Paris, France. 5 Department of health sciences, University of Genova, Genova, Italy. 6 Dipartimento di Radioterapia, Ospedale Policlinico San Martino, Genova, Italy. 7 CentraleSupélec, Mathématiques et Informatique pour la complexité et les systèmes, Université Paris Saclay, Gif-Sur-Yvette, France Purpose/Objective: This study aimed to evaluate the predictive power of advanced perfusion biomarkers, including Capillary Transit Time Heterogeneity (CTH) and Oxygen Extraction Fraction (OEF), for identifying recurrence volume (L3R) site on pre radiotherapy (pre-RT) MRI. Material/Methods: We analyzed data from patients with high-grade glioma who underwent pre-RT MRI with DSC perfusion sequences in a single institution. Patient data included demographics, molecular and surgical details. Residual disease volume was assessed on pre-RT MRI. Perfusion maps (rCBV, CTH, and OEF) were generated using the Cercare Perfusion solution (Aarhus , Denmark). rCBV was used as the reference standard. Tumor regions were segmented into five labels on pre-RT and relapse MRI: necrotic core (Label 1), T2/FLAIR hyperintensity (Label 2), contrast-enhancing tumor (Label 3), surgical cavity (Label 4), and postoperative modifications (Label 5). A rigid image registration was applied to align the relapse images with pre-RT images. For each patient and biomarker, we plotted ROC curves, calculated AUC, and determined the optimal classification threshold by Youden Index. ROC analysis was performed considering all voxels in the population. The method of analysis involved a voxel-based comparison of relapse volume in the pre-RT imaging. Results: We included 197 patients with a median age of 57.6 years(18–84yrs). The cohort comprised 80 patients IDHwt, 14 IDHm, and 103 IDHstatus-unknown. A total of 183 patients underwent the 'Stupp’ protocol. Residual disease was observed in 134 patients. Relapse was observed in the whole cohort, with a median Progression-Free Survival of 14.8 months and an OS of 26.3 months. Recurrence patterns included local (63.5%), distant (15.2%), and synchronous (21.3%) relapses. In the ROC analysis for L3R prediction, CTH demonstrated an AUC of 0.69 ± 0.16, with sensitivity and specificity values of 0.604 and 0.689, using a cutoff of 2.35. rCBV demonstrated a sensitivity and specificity of 0.959 and 0.169, with an AUC of 0.58 ± 0.16 (Figure 1).