TMOD-36. Quantifying Tumor Kinetics in Glioblastoma via Reaction-Diffusion Modeling Across Multimodal Treatment Phases using Longitudinal MRI images

Glioblastoma (GBM) is a highly invasive brain tumor characterized by complex, patient-specific growth patterns. Capturing these dynamics longitudinally can improve treatment response understanding and guiding personalized therapy. We present a reaction diffusion modeling framework to estimate tumor proliferation, diffusion, and treatment response from longitudinal MRI data. We analyzed a GBM patient (IDHwildtype, OS=1601 days) with 20 longitudinal T1C scans. CE tumor masks were manually segmented, and SPM-derived gray and white matter, and CSF maps were combined at each timepoint to guide spatially varying diffusion simulation. For each interval, the previous tumor mask served as the initial seed, and the difference with the next timepoint defined the final seed, capturing both progression and regression. A reaction diffusion model was used to simulate tumor evolution, incorporating proliferation and a kill term to reflect treatment effects. Patient-specific parameters (diffusivity in gray/white matter, proliferation, and kill rate) were optimized using Bayesian optimization to minimize the average surface distance (ASD) between simulated and observed tumor progression. During early treatment with radiotherapy and temozolomide (TMZ), the tumor exhibited marked regression in contrast-enhancing volume. This was captured by the model through strong treatment associated kill effects (~0.71–0.77 per day) and moderate diffusivity in gray and white matter (0.13–0.44 and 0.20–0.72 mm²/day, respectively). In later stages, under continued TMZ, bevacizumab, and radiotherapy, the tumor transitioned to a more infiltrative pattern, reflected by rising white matter diffusivity (~0.95 mm²/day), while proliferation remained consistently low (0.02–0.04 per day). Simulated tumor load maps closely matched follow-up segmentations, achieving ASD errors below 2 mm across most intervals. This framework captures patient-specific tumor dynamics and treatment response. Increasing patients could enable population-level modeling to better understand disease progression and guide personalized therapy.