Pre-Surgical Detection of Infiltrating Glioblastoma Using a Histopathology-Validated MR Fingerprinting Prediction Model

Motivation: Glioblastoma (GBM) infiltration models lack histopathological validation. Goal(s): GBM infiltration models should be validated with histopathological data of the surrounding peritumoral zone (PZ), which is typically omitted from surgery and difficult to obtain retrospectively. Approach: We prospectively obtained histopathology data from either targeted biopsies (n = 5) or extended peritumoral resection (n = 18). GBM cases with ground truth PZ sampling were used to evaluate the performance of a self-supervised MR fingerprinting (MRF) and multiparametric MRI classifier. Results: Our histopathology-validated infiltration model achieved 70.9% sensitivity and 74.5% specificity in detecting infiltrating glioma, with a mean balanced accuracy of 72.7%. Impact: Glioblastoma (GBM) peritumoral infiltration leads to inevitable recurrence and death. We develop and histopathologically-validate an MRF artificial intelligence (AI) model for pre-surgical prediction of infiltrating GBM to reduce tumor recurrence, improve patient quality of life, and extend survival.