Fig 1.

Radiomics workflow. The radiomic clustering workflow begins with the feature extraction pipeline, depicted in the rectangle at the left of the schematic. Patients first undergo image acquisition through routine MR imaging protocols. Images are then accessed and pre-processed which involves stripping away extra-axial tissues from the images leaving only the brain and tumor volumes. The tumor is then segmented out leaving only tumor volume. Engineered radiomic features, including texture-based, morphologic, and histogram features, are then extracted from the tumor volumes. The features are then fed into the unsupervised machine learning algorithm, which sorts the patients based on the radiomic features. The end results are distinct groups of patients, categorized based on imaging, that can be compared on their clinical, genetic, and other molecular characteristics.