KEY POINTS
- A diffusion-proliferation model was applied to a 157-patient glioblastoma imaging cohort, but treatment planning was explored in only three selected recurrent cases representing different tumor locations and infiltration patterns.
- Instead of a conventional isotropic CTV margin, the model estimated anisotropic microscopic spread along brain anatomy. Target volumes were defined using simulated cell-density thresholds of 1000, 100, and 10 cells/mm³, with a 3-mm PTV expansion.
- All model-derived target plans used 60 Gy in 30 fractions and met the predefined clinical planning criteria: D95 >95% of prescription and D2 <107%, while organ-at-risk limits were maintained.
- Predicted microscopic disease substantially enlarged the treatment volume. CTV1000 volumes were 57.7, 122.7, and 42.8 cm³ in the three cases, compared with GTVs of 24.2, 41.9, and 14.5 cm³, respectively.
- Across the anisotropic plans, mean homogeneity index was 0.90 and mean conformity index was 0.93, suggesting that even irregular model-derived targets could be treated with acceptable conventional VMAT geometry.
- A second strategy prescribed dose voxel-by-voxel according to predicted tumor-cell density and a theoretical tumor-control probability of 0.95. Dose mimicking was most accurate inside the GTV, with mean Q0.95–1.07 of approximately 95%, but deteriorated as the evaluated volume increased.
- The dose-painting approach also produced variable normal-brain exposure: healthy-brain volume receiving ≥60 Gy was 60.61, 4.84, and 0.28 cm³ across the three cases. This highlights the central trade-off between treating modeled infiltration and avoiding normal-brain dose.
CLINICAL TAKEAWAY
Model-informed glioblastoma target definition is technically feasible and could eventually offer a more biologically plausible alternative to uniform isotropic CTV margins. At present, however, this is a planning proof of concept rather than evidence that targeting predicted microscopic spread improves local control or survival.