Time-dependent diffusion MRI versus conventional DWI for dual-parameter assessment of non-enhancing gliomas: IDH mutation and Ki-67 labeling index

Noninvasive assessment of isocitrate dehydrogenase (IDH) mutation status and Ki-67 labeling index in non-enhancing gliomas is challenging. This study aimed to compare the performance of time-dependent diffusion MRI (TDD-MRI) with that of conventional diffusion-weighted imaging (DWI) in classifying IDH mutation status and Ki-67 labeling index in non-enhancing gliomas. In this single-center retrospective study, 173 patients were chronologically divided into development ( n = 99) and temporal validation ( n = 74) cohorts and stratified by IDH mutation status and Ki-67 index into three subgroups: IDH wild type (IDHwt), IDH-mutant with low Ki-67 (IDHmut & Ki-67_low), and IDH-mutant with high Ki-67 (IDHmut & Ki-67_high). Histogram features extracted from seven TDD-MRI-derived parameter maps and a conventional DWI-derived apparent diffusion coefficient (ADC) map were selected using least absolute shrinkage and selection operator regression and Boruta and compared across the three subgroups. Conventional ADC (cADC) models, uni-parametric histogram models (UPMs), and multi-parametric histogram models (MPMs) were then constructed for three one-versus-rest binary classification tasks. The probability generated by the best-performing model was incorporated into a Clinical model comprising age and tumor location to form a Clinical + TDD model, and all models were evaluated. The development cohort comprised 18 IDHwt, 57 IDHmut & Ki-67_low, and 24 IDHmut & Ki-67_high gliomas, whereas the temporal validation cohort comprised 15, 39, and 20 gliomas, respectively. Several TDD-MRI and cADC histogram features differed significantly among the three subgroups in both cohorts. Among the cADC, UPM, and MPM models, the MPM yielded the highest AUC: 0.892, 0.852, and 0.836 in the development cohort and 0.875, 0.809, and 0.825 in the temporal validation cohort, respectively. When its probability was incorporated into the Clinical model, the Clinical + TDD model yielded numerically higher AUCs than the Clinical model for identifying IDHwt, IDHmut & Ki-67_low, and IDHmut & Ki-67_high, with AUCs of 0.929, 0.858, and 0.836 versus 0.789, 0.689, and 0.636 in the development cohort and 0.951, 0.830, and 0.828 versus 0.860, 0.693, and 0.644 in the temporal validation cohort, respectively. TDD-MRI showed preliminary value over conventional DWI for discriminating prespecified IDH/Ki-67 strata of non-enhancing gliomas and provided additional discriminative information beyond clinical variables.