Can parametric response maps predict voxel-wise treatment response? Implications for locally adaptive radiotherapy

Abstract

Parametric response map (PRM) analysis is a voxel-based image analysis method for predicting treatment response which shows promise as a means for guiding locally adaptive radiotherapy (RT). However to date, studies have focused on verifying PRM predictive utility with respect to global outcomes such as overall survival (OS). Here we investigated whether voxel-wise treatment response information can be inferred from a PRM analysis that has been correlated with OS. PRMs were generated from repeat MRI-derived apparent diffusion coefficient (ADC) maps (1 and 3 months post-RT) for n = 14 patients treated for high-grade glioblastoma. The proportion of voxels in each PRM class that remained within the tumor boundary at 6 months post-RT was computed. Voxels classified as significantly increasing in ADC were more likely to remain within the boundaries of the tumor at 6 months post- RT compared to voxels classified as significantly decreasing in ADC (p < 0.001). However in contrast, the fractional tumor volume classified as significantly increasing in ADC was positively correlated with OS (ρ = 0.63, p = 0.02). The PRM was found to show potential for predicting both global and voxelwise treatment response, however, the relationship between the two could not be directly inferred suggesting that rigorous validation is needed if the PRM is to be used to guide locally adaptive RT.