Related Experiment Video
Updated: Sep 26, 2026
Positron Emission Tomography-based Dose Painting Radiation Therapy in a Glioblastoma Rat Model using the Small Animal Radiation Research Platform
Published on: March 24, 2022
Patient-Specific Computational Framework to Spatiotemporally Forecast Treatment Response and Identify Biologically
Sophia Ty1, Jeanne Kowalski2,3, Bikash Panthi4
1Oden Institute for Computational Engineering and Sciences, The University of Texas at Austin, Austin, TX.
Purpose:
High-grade gliomas exhibit substantial spatial and temporal heterogeneity, posing a challenge for radiotherapy (RT) target definition. Adaptive RT, which prospectively modifies treatment plans, may benefit from early identification of intratumoral regions likely to exhibit differential treatment response. Apparent diffusion coefficient (ADC), a marker associated with tumor cellularity and aggressiveness, is a promising imaging biomarker of RT response. We present a patient-specific computational framework that predicts treatment response and identifies candidate adaptive RT targets from anticipated spatiotemporal patterns of ADC changes (ΔADC).
Methods:
Using longitudinal multiparametric magnetic resonance imaging (MRI) from 21 patients, the mathematical model was calibrated to spatiotemporally forecast tumor cell density immediately after and 1-month post-RT. Predictions of tumor cell density were converted to ADC maps. ΔADC from baseline was analyzed using Pareto tail analysis. The lower ΔADC distribution tail was used to classify tumor voxels as regions of increased diffusion restriction (IDR) or decreased/maintained diffusion restriction (DMDR). We compared agreement in categories between measured and predicted ΔADC from MRI data. Classification performance was assessed using receiver operating characteristic (ROC) analysis across four scenarios combining two time points and two image inclusion strategies (± nonenhancing tumor volume).
Results:
The framework achieved a median ROC AUC of 0.88 (0.80-0.93), indicating strong performance in distinguishing IDR from DMDR regions. No significant differences were observed between forecasting time points (P value: .08) or image inclusion strategies (P value: .50).
Conclusion:
This framework can delineate IDR and DMDR regions, offering a potential strategy for prospectively defining candidate biologically informed adaptive RT targets. Future studies are needed to evaluate their role in adaptive RT planning.
