Machine learning-supported MRI Radiomics to predict the volumetric response in meningiomas after Gamma Knife radiosurgery.

Abstract

Background: In previous studies, we analyzed the potential of both Diffusion Tensor Imaging and of Texture Analysis of Magnetic Resonance Imaging to predict the volumetric response of benign meningiomas to Gamma Knife radiosurgery (GKRS). In this study, we analyzed the value of meningioma morphology in the prediction of volumetric changes induced by GKRS. Methods: The retrospective prediction model of meningioma responsiveness to GKRS included T1- weighted, non-contrast enhanced MRI scans obtained from 93 patients before GKRS. Imaging data was processed and analyzed through the QMENTA cloud platform and meningioma morphology was quantified by calculation of 337 shape, first-order and second order radiomic features. This analysis was performed on original 3D unfiltered MR images and images modified by Laplacian of Gaussian (LoG), logarithm and exponential filters. Results: Sixty calculated features significantly correlated with the outcome defined as meningioma volume change per month. The predictive model was created based on all radiomic and twelve non-radiomic features using the LASSO regression machine learning method. Thereby, LoG-sigma-1-0-mm-3D_firstorder_InterquartileRange (coefficient weight = -9.916) and logarithm_ngtdm_ Busyness (coefficient = 0.002) were selected as the most predictively robust and non-redundant features. The radiomic score based on these two radiomic features produced an AUC = 0.81. Its values ranged between -2.89 and 2.48, whereby score values up to -1.31 defined a subgroup of 50 patients with consistent absence (0%) of tumor progression. Conclusions: This is the first report of a strong association between the MRI radiomic features and the volumetric meningioma response to radiosurgery. The clinical importance of the early and reliable prediction of meningioma responsiveness to GKRS is based on its potential to guide individualized treatment strategies.