Combined Use of APT-MRI and VASARI Features for Assessing the Aggressiveness of Diffuse Gliomas

Objective . This study aimed to conduct a comprehensive analysis of the potential of combined use of Amide Proton Transfer (APT) imaging and standardized VASARI (Visually AcceSAble Rembrandt Images) features for predicting molecular characteristics (IDH1 status, 1p19q codeletion) and the grade (differentiating Grade 3 from Grade 4) in patients with diffuse gliomas. Materials and Methods . A prospective pilot study included 40 patients with morphologically verified diffuse gliomas. All patients underwent preoperative MRI on a 3T scanner, including an APT sequence, as well as ASL and DWI (with ADC calculation). Quantitative assessment of APT, ASL, and ADC was performed, along with a qualitative analysis of 25 morphological features according to the VASARI system, followed by multiple correspondence analysis to identify latent components. Machine learning methods (Random Forest, XGBoost, Logistic Regression, Support Vector Machine, Multinomial Regression) were applied for classification tasks. The performance of models based solely on quantitative MRI parameters (mean APT, ADC, ASL values) was compared with models that additionally included VASARI components. Results . APT values significantly differed between gliomas of different grades (p = 0.004), while ASL and ADC showed no significant differences. The extracted VASARI components explained 37.2 % of the variance, with the first and most significant component (VASARI MC1), reflecting the tumor's invasive potential, demonstrating a strong correlation with APT (r = 0.65, p < 0.001). The best results were achieved in predicting IDH1 status: a Random Forest model incorporating VASARI features showed an accuracy = 0.9 and AUC = 0.91, which was significantly higher than models based solely on quantitative parameters (accuracy 0.8, AUC 0.81). Predicting the presence/absence of 1p19q codeletion and differentiating Grade 3 from Grade 4 tumors proved to be more challenging tasks, and the addition of VASARI did not lead to a significant improvement in the models. Conclusion . The combined use of quantitative APT parameters and standardized visual assessment using the VASARI system demonstrates significant potential for the non-invasive prediction of molecular-genetic characteristics of diffuse gliomas, particularly IDH1 status. The integration of qualitative and quantitative MRI data allows for the creation of more accurate diagnostic models, which may contribute to the personalization of patient treatment strategies.