Voxelwise Mapping of Whole-Exome Sequencing to Noninvasive Imaging with Stereotactic Localization and Deep Learning

Abstract

Tumor heterogeneity presents a significant barrier to glioblastoma treatment. While multiparametric imaging can improve characterization of such heterogeneity, advances have been limited by access to ground truth pathology and genomics. Stereotactic localization addresses the former issue, allowing pairing of biopsy location with corresponding voxels in MR image. Using stereotactic localization, as well as voxelwise MR image analysis, a feedforward dense network, and whole-exome sequencing, the authors sought to improve multiparametric MR characterization of tumor heterogeneity.