Noninvasive Prediction of TP53 Gene Status and ATRX Gene Status in IDH-Mutant Glioma Using Multimodal MRI:...

Background/Objectives: Noninvasive determination of glioma molecular profiles is clinically crucial for assessing therapeutic efficacy and predicting disease outcomes. This study aimed to evaluate the potential of morphological magnetic resonance imaging (MRI), diffusion-weighted imaging (DWI), magnetic resonance spectroscopy (MRS), and dynamic contrast-enhanced perfusion-weighted imaging (DCE-PWI) in predicting TP53 gene status and X-linked alpha-thalassemia intellectual disability syndrome (ATRX) gene status in isocitrate dehydrogenase (IDH)-mutant gliomas. Methods: A retrospective analysis was performed on 106 IDH-mutant glioma patients using morphological MRI, DWI, MRS, and DCE-PWI data. Statistical comparisons of imaging parameters across molecular status groups were conducted, and logistic regression models were developed to predict molecular status, with diagnostic performance evaluated by receiver operating characteristic (ROC) curve analysis. Five-fold stratified cross-validation with 1000 bootstrap resamples was employed to assess model generalizability Results: Among 106 IDH-mutant gliomas, the TP53-mutant group showed a greater proportion of tumors with >33% enhancement (p = 0.018), higher Cho/Cr (p < 0.001), and higher Cho/NAA (p = 0.005) than the TP53-wildtype group. Multivariable analysis demonstrated that the Cho/Cr ratio was an independent predictor of TP53 mutation in IDH-mutant gliomas (odds ratio [OR] = 2.037, p = 0.021), with the model achieving an apparent AUC of 0.741. DCE-PWI parameters showed no significant differences across molecular subgroups. Ve was significantly elevated in ATRX-mutant tumors (median 57.16 vs. 30.63, p = 0.029). Ktrans, Kep, Vp, and iAUC showed no significant differences between groups (all p > 0.05). Furthermore, multivariable analysis showed that ADC values (OR = 1.005, p = 0.017) and the Cho/NAA ratio (OR = 3.073, p = 0.023) emerged as independent predictors of ATRX mutation, with the model achieving an apparent AUC of 0.863. Five-fold cross-validation demonstrated that the Cho/Cr model for TP53 prediction achieved a mean AUC of 0.717 ± 0.043 (Bootstrap 95% CI: 0.616-0.814), and the ADC + ChoNAA model for ATRX prediction achieved 0.865 ± 0.124 (95% CI: 0.780-0.953). All predictors remained significant across all five folds. Pooled confusion matrices yielded sensitivities of 0.623 and 0.757, specificities of 0.696 and 0.909, and accuracies of 0.654 and 0.840, respectively. Conclusions: Multimodal MRI techniques (morphological MRI, DWI, MRS, and DCE-PWI) can help predict TP53 and ATRX status without surgery. Higher Cho/Cr and Cho/NAA ratios were independently associated with TP53 mutation, whereas lower ADC and higher Cho/NAA independently predicted ATRX mutation. These findings suggest that a focused imaging protocol may be sufficient for preoperative molecular profiling in this tumor type.