Fig. 1.

The workflow of Radiomics in this study is graphically presented in four steps. (1) T2-weighted MR images were collected from four different centers in the Netherlands and Belgium and manually segmented. The MR volumes and their corresponding segmentation masks were preprocessed into isotropic voxels. (2) Four types of features (a. Shape features, b. First-order statistic features, c. Texture features, and d. Features extracted after applying different filters) were extracted from the segmented masks. (3) Feature reduction was done by principal component analysis. (4) A multi-layer perceptron classifier was used for radiomic analysis