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

摘要: The purpose of this study was to develop a voxel-wise clustering method multiparametric magnetic resonance imaging (MRI) and 3,4-dihydroxy-6-[18F]-fluoro-L-phenylalanine (FDOPA) positron emission tomography (PET) images using an unsupervised, two-level approach followed by support vector machine in order classify the isocitrate dehydrogenase (IDH) status gliomas. Sixty-two treatment-naive glioma patients who underwent FDOPA PET MRI were retrospectively included. Contrast enhanced T1-weighted images, T2-weighted fluid-attenuated inversion recovery apparent diffusion coefficient maps, relative cerebral blood volume used for feature extraction. An unsupervised approach, including self-organizing map K-means algorithm used, each class label applied original images. logarithmic ratio labels within tumor regions differentiate IDH mutation status. area under curve (AUC) receiver operating characteristic curves, accuracy, F1-socore calculated as metrics performance. associations values cluster successfully visualized. Multiparametric with 16-class revealed highest classification performance AUC, F1-score 0.81, 0.76, respectively. Machine learning classified gliomas, visualized features from Unsupervised clustered may improve understanding prioritizing classifying