Time-dependent diffusion MRI-based microstructural mapping for predicting IDH mutation status in glioma: A...

To evaluate whether time-dependent diffusion MRI(td-dMRI)-based microstructural histogram parameters can accurately distinguish IDH mutation status in gliomas, with multicenter external validation.

In this prospective dual-center study, patients with pathologically confirmed glioma underwent preoperative conventional MRI and td-dMRI-based. Microstructural parameters including intracellular volume fraction (Vin), intracellular diffusivity (Din), extracellular diffusivity (Dex), cellularity, and cell diameter-were derived using the IMPULSED model. ADC values at different frequencies (25 Hz, 50 Hz, PGSE) were also computed. Tumor regions were manually segmented by two radiologists, excluding necrosis and edema. Voxel-wise values were extracted, and histogram features were computed using PyRadiomics. Feature selection included the Mann-Whitney U test, Spearman correlation, and logistic regression. A predictive model for IDH mutation status was developed using data from Center A and externally validated in Center B. Model performance was assessed using AUC, calibration curves, and confusion matrices.

Among 147 histogram features extracted from seven td-dMRI-based mapping, eight nonredundant features significantly differed between IDH-wildtype and IDH-mutant gliomas. A logistic regression model based on ADC_PGSE_firstorder_Energy and cellularity_firstorder_10Percentile yielded AUCs of 0.801 (training) and 0.771 (validation). ADC_PGSE and cellularity were positively correlated with the Ki-67 index (P < 0.05).

Histogram features from td-dMRI (ADC_PGSE and cellularity) enabled robust IDH mutation prediction in a multicenter glioma cohort.