2692 - Improving Glioblastoma Radiotherapy Target Delineation via a Diffusion Model

08:00am - 09:00am PT

Hall F

Screen: 20

POSTER

Presenter(s)

Wan-Jin Yeo, PhD - University of California, San Francisco, San Francisco, CA

W. J. Yeo¹, S. Nagarajan¹, A. Raj¹, D. Ruan², W. Yang¹, and K. Sheng³; ¹University of California, San Francisco, San Francisco, CA, ²Department of Radiation Oncology, University of California, Los Angeles, Los Angeles, CA, ³Department of Radiation Oncology, University of California, San Francisco, San Francisco, CA

Purpose/Objective(s):

The standard clinical target volume (CTV) for glioblastoma (GBM) radiotherapy (RT) is a 2 cm isotropic margin defined from the pre-operative gross tumor volume, but this uniform approach may not be an effective treatment plan for all patients. Previous studies have shown that GBM predominantly recurs locally and may invade along white matter (WM) tracts. In this study, we aim to improve individualized CTV delineation via predicting GBM recurrent volume through a diffusion model, informed by diffusion tensor (DT) imaging.

Materials/Methods:

The study population was 35 patients from Río Hortega University Hospital (mean age 63.2 years, 28.9% women) with locally-recurrent GBM. GBM was defined with the necrotic, enhancing, and non-enhancing regions segmented from their T1, T1ce, and T2-FLAIR magnetic resonance (MR) images. For each patient, an anisotropic 3D reaction-diffusion equation solved numerically with an alternating-direction implicit scheme in voxel space was implemented to predict GBM growth. The model was subjected to the Neumann boundary condition at the skull and had initial condition being the pre-operative tumor. Diffusivity was modeled with the diffusion tensor obtained from DT imaging. We assessed the performance of our model’s boundary in predicting the recurrent GBM against the 2 cm isotropically-expanded boundary by evaluating, in both cases, the proportion of recurrent tumor encapsulated per increase in prediction tumor volume.

Results:

The mean proportion of recurrent tumor captured by the 2 cm boundary was 57% (standard deviation (SD) 24.9%), whereas with our predicted boundary at the same volume, a mean of 58.4% (SD 24.9%) of the recurrent tumor could be captured. In 26 out of 35 patients (74.3%), the boundary predicted by our diffusion model was able to capture an equivalent volume of the recurrent tumor as the 2 cm isotropic boundary with a 1.2% to 30.4% decreased prediction volume.

Conclusion:

Relative to isotropically-expanding the boundary of an initial GBM, a diffusion model informed by patient-specific diffusivity can have an improved prediction of recurrent GBM growth with a smaller prediction volume. This implies that GBM RT may be more personalized, and CTVs may be improved by reducing the treatment volume without compromising treatment quality.