Towards predicting tumor pathology with radiopathomic analysis of multi-parametric MRI in patients with newly-diagnosed gliomas

Motivation: Noninvasive identification of malignant regions in glioma can help guide diagnosis and subsequent treatment planning. Goal(s): This study aims to create models to predict and elucidate limitations in radiopathomic mapping of invasiveness in glioma using multiparametric physiologic and metabolic MRI. Approach: A large, unique multiparametric MRI dataset with tissue is leveraged to compare various machine learning models of %ki-67 and cellularity (cells/mm2). Results: : The best binary model achieved a CV-AUC =0.82 and CV-AUC = 0.75 for a binarized ki-67 and cellularity. Best ki-67 continuous predictions were in the 10-fold CV SVM and 4-fold ensemble model for continuous cellularity. Impact: Multiparametric MRI can non-invasively predict histopathology. Including physiologic and/or metabolic MRI boosts histopathological predictions, however performance is also impacted by standardization of data quality.