Figure 1.

The radiomics workflow of the study. (a) PET image with lesion segmentation based on the thresholding method (see Section 2.1). (b) Some examples of the 44 features extracted using LifeX software, such as kurtosis, skewness, entropy and two textural features in the form of matrix (see Section 2.2), (c) Feature selection using a mixed descriptive-inferential sequential approach (see Section 2.2.3), (d) Representative figure of biplanar feature classification (low and high tumour grade) using the discriminant analysis as the supervised classifier. Specifically, a set of labelled instances (low and high tumour grade) were used for training purpose. Successively, the classifier was used to predict the label in the test set, where instances were without corresponding labels (see Section 2.2.4).