Figure 1.

Work flow and radiomic feature screening. (A) Work flow of the machine learning method. (B) A total of 1293 RFs showed high intraclass correlation coefficient (ICC) between tumours sketched by radiologist or neurosurgeon. (C) The frequency of RFs chosen (univariate Cox model, likelihood ratio test, P < 0.05, both in the feature value and the median of RF score) in the prediction models. (The peaks are in the 850 times.) (D) The RFs of Kaplan–Meier P-values in the test groups and frequency. OS = overall survival; ROIs = regions of interest.