Differentiating IDH status in human gliomas using machine learning and multiparametric MR/PET

Area of Science:

• Neuro-oncology
• Medical imaging analysis
• Machine learning in medicine

Background:

• Gliomas are brain tumors where isocitrate dehydrogenase (IDH) mutation status is a critical prognostic and predictive biomarker.
• Accurate IDH status classification is essential for guiding glioma treatment strategies.
• Current methods for determining IDH status can be invasive or time-consuming.

Purpose of the Study:

• To develop and validate a novel voxel-wise clustering method integrating multiparametric magnetic resonance imaging (MRI) and 3,4-dihydroxy-6-[¹⁸F]-fluoro-L-phenylalanine (FDOPA) positron emission tomography (PET) imaging.
• To classify the isocitrate dehydrogenase (IDH) mutation status of gliomas using an unsupervised, two-level clustering approach followed by a support vector machine (SVM).
• To visualize voxel-wise features from multiparametric MRI and FDOPA PET images to enhance understanding of imaging biomarkers for IDH status.

Main Methods:

• Retrospective analysis of 62 treatment-naïve glioma patients who underwent both FDOPA PET and MRI.
• Extraction of voxel-wise features from contrast-enhanced T1-weighted MRI, T2-weighted MRI, fluid-attenuated inversion recovery (FLAIR) MRI, apparent diffusion coefficient (ADC) maps, relative cerebral blood volume (rCBV) maps, and FDOPA PET images.
• Application of an unsupervised two-level clustering approach (Self-Organizing Map followed by K-means) and subsequent classification using a Support Vector Machine (SVM) with logarithmic ratio of class labels.

Main Results:

• Successful visualization of associations between multiparametric imaging values within identified clusters.
• The 16-class clustering approach demonstrated the highest performance in differentiating IDH status.
• Achieved an Area Under the Curve (AUC) of 0.81, accuracy of 0.76, and F1-score of 0.76 for IDH status classification.

Conclusions:

• Machine learning, combining unsupervised clustering and SVM, effectively classifies IDH mutation status in gliomas using multiparametric MRI and FDOPA PET.
• The developed method provides visualized voxel-wise features, potentially improving the interpretability of imaging data.
• Unsupervised clustered features offer insights into prioritizing specific multiparametric imaging modalities for accurate IDH status determination in gliomas.