Room: AAPM ePoster Library
Purpose: To develop a spatiotemporal model that can be calibrated with subject-specific imaging measures to predict individual tumor response to fractionated radiation therapy.
Methods: Quantitative magnetic resonance imaging (MRI) data was collected in two rats with intracranially injected U87 glioblastoma cells before, during, and following 10 fractionated doses of 2 Gy. At each visit, diffusion weighted (DW-) and dynamic contrast-enhanced (DCE-) MRI were collected to measure tumor and blood volume fractions, respectively, in 3D. We developed a coupled system of two partial differential equations describing the diffusion, proliferation, and death of tumor cells and vasculature due to radiation therapy. We evaluated two models of response to radiation therapy. Model 1 assumes radiation therapy introduces both a reduction in the fraction of proliferation tumor cells and an increase in the fraction of tumor cells that are dying. Model 2 assumes an immediate reduction of tumor cells similar to the linear quadratic model. The coupled model was implemented and solved using a fully explicit in time, finite difference approximation. Animal specific model parameters were calibrated using the pre- and early treatment data. The calibrated model parameters were then used to predict future tumor growth. Prediction error was assessed at the global (percent error in total tumor volume) and local (percent error in voxel-wise estimates of model outputs) levels.
Results: For model 1, percent error in tumor volume ranged from 1.3 to 14.1% while the average percent error at the voxel-level was < 15.5% For model 2, percent error in tumor volume ranged from 6.9 to 589% while the average percent error at the voxel-level was less than 21.0%.
Conclusion: Quantitative imaging data can be used to calibrate a predictive mathematical model of response to fractionated therapy and, importantly, a linear-quadratic characterization of radiation induced cell death may not be optimal.
Funding Support, Disclosures, and Conflict of Interest: This work was supported through funding from the National Cancer Institute R01CA138599, U01CA174706, CPRIT RR160005, and AAPM Research Seed Funding. The authors acknowledge the Texas Advanced Computing Center for providing high-performance computing resources. T.E.Y. is a CPRIT Scholar of Cancer Research.
TH- Radiobiology(RBio)/Biology(Bio): Rbio - Outcome models combining dose, imaging, radiomics/radiogenomics and clinical factors: other than machine learning