Fig. 4

Confusion matrices showing the performance of the classifiers based on (a) only clinical data and (b) the combination of clinical data with radiomic features. Columns indicate the number of subjects who actually belong to the class, while rows indicate the number of subjects predicted to belong to the class. The percentages around the edge of the matrices refer to the accuracy (green text) and error rates (red text) of each column, each row, and the whole matrices (lower right corner)