2655 - Planning Assessment of Pre-RT White Matter Mapping-Enhanced CTV for Glioblastoma Radiation Therapy

Purpose/Objective(s): Conventional RT approaches for glioblastoma employ a clinical target volume (CTV), delineated by expanding the atomical lesion visible on T1 contrast-enhanced (T1ce) and T2-weighted FLAIR MRI, encompassing the resection cavity and residual contrast-enhancing lesions, plus an isotropic 1-2 cm margin to account for subclinical spread. Despite aggressive multimodal treatment, GBM recurs in up to half of tumors with failure outside of the CTV, often along white matter tracts. To address this, we have proposed a personalized CTV, generated by a deep learning model incorporating anatomical MRI and the fiber density-weighted white matter pathlength (DW-WMPL) mapping. We hypothesize that RT plans generated using this DW-WMPL-enhanced CTV will provide better coverage to areas of recurrence while simultaneously reducing the dose to organs at risk.

Materials/Methods: In prior work, a 3D SWINUNETR deep learning model was trained to predict areas of GBM recurrence from post-operative T1ce, T2 FLAIR and DW-WMPL map in a cohort of 125 patients. A holdout test set of 12 patients was used to generate ground truth progression masks from post-RT follow-up imaging via auto-contouring, which were validated by clinicians. For each patient in the test set, two de novo RT plans were created: one using the clinical CTV, and another with the DW-WMPL-enhanced CTV. Both plans delivered 6000 cGy to at least 95% of the target volume, while meeting NRG dose constraints for organs at risk. All plans were created for a C-arm linac and dose was computed using collapsed cone convolution at a 2mm isotropic resolution. Dose statistics were extracted for each plan and compared in a pairwise manner using the Wilcoxon signed-rank test.

Results: Treatment plans using the DW-WMPL-enhanced CTV increased median prescription coverage of disease progression by 21.5% (67.5% vs. 88.9%, p = 0.034) compared to the clinical CTV. Although not statistically significant, the DW-WMPL plans showed numerically lower maximum point doses (D0.03 cc) to the brainstem, optic chiasm, and optic nerves, and reduced 80% isodose and CTV volumes by 16.7% and 11.6%, respectively.

Conclusion: Personalization of the GBM CTV with a DW-WMPL mapping approach can significantly improve dose coverage of future disease progression, while lowering doses to critical structures. This approach shows promise for enhancing therapeutic outcomes for GBM and warrants further investigation in larger cohorts.

Abstract 2655 - Table 1: Dose statistics for the clinical CTV and DW-WMPL-enhanced CTV, reported as median (range)