Histogram analysis of IVIM MRI for differentiating glioma IDH-1 status, grade, and Ki-67 expression: a comparison of bi-exponential models at multiple B-value

Intravoxel Incoherent Motion (IVIM) offers valuable insights into tumor microstructural heterogeneity through diffusion and perfusion. However, the relationship between these features and glioma grading, IDH-1 status, and Ki-67 expression remains insufficiently explored, and there is no consensus on the optimal B-value range for IVIM imaging, which may impact the consistency of results. We retrospectively analyzed IVIM data from 73 glioma patients (mean age, 52.83 years ± 12.79 [SD]; 40 male). Histogram features of IVIM parameters, calculated with four B-value ranges (0-800, 0-1200, 0-1500, and 0–2000 s/mm²), were extracted from the tumor parenchyma. These features were compared across different IDH-1 status, tumor grade, and Ki-67 proliferation index. After checking for multicollinearity, parameters with significant differences were further analyzed using logistic regression, receiver operating characteristic analysis, and decision curve analysis to compare their performance. And Bonferroni correction was applied to adjust p-values for multiple comparisons. Independent risk factors of IDH-1 status, tumor grade and Ki-67 level were identified through multivariate logistic regression in B800, B1200, B1500 and B2000 model. ROC analysis revealed that all models showed similar AUCs for IDH-1 status, tumor grade, and Ki-67 level prediction. The B800 model had the highest specificity (0.87) for IDH-1 status prediction, the B1500 model had the highest specificity (0.91) for tumor grade prediction, and the B1200 model exhibited the highest sensitivity (0.94) for Ki-67 level prediction. The B1500 and B 2000 model demonstrates the best net benefit in most threshold ranges for IDH-1 status, tumor grade and Ki-67 level identification. Histogram features of IVIM parameters demonstrated well performance in distinguishing glioma characteristics and the maximum 1500 s/mm2 may be the optimal.