Longitudinal voxel-wise analysis using a novel deep-learning-derived KI-67 map for early prediction of glioblastoma outcome

Motivation: Early prediction of prognosis, treatment response, and overall-survival is imperative for adapting appropriate treatment strategies for GBM patients. Goal(s): To predicting outcome and tumor progression using early imaging changes in underlying tumor pathology Approach: We performed a longitudinal voxel-based analysis of metabolic and proliferation changes during the course of radiation therapy in terms of their ability to predict progression and survival compared to anatomical tumor volumes. Results: Voxel-wise subtraction (MidRT-PreRT) of proliferation values was the most correlated with progression free (r=0.54/p=0.004) and overall (r=0.66/p=0.0002) survival, exceeding both tumor volume and CNI. Adding imaging-derived metrics of proliferation mid-RT can improve the prediction of progression. Impact: Novel imaging-based proliferation maps and voxel-wise analyses show much stronger correlation with progression-free and overall-survival and regions of subsequent progression compared to conventional imaging-markers. These features can potentially aid in early prediction of response, and outcome of patients with GBM.