Fig. 3: Results in internal cross-validation and hold-out testing cohort from the different methods.
From: Deep mutual learning on hybrid amino acid PET predicts H3K27M mutations in midline gliomas
A–D Internal cross-validation and E–G Hold-out testing cohort. A, B Boxplots for the slice-level results from MET and FET, respectively. C, D Bar plots for the individual-level results from MET and FET, respectively. E, F Corresponding slice-level results for A, B in the hold-out testing cohort, as bar plots. G Bar plots for the individual-level results from combined MET and FET data in the hold-out testing cohort. * Results from CNN with assistance training (AT-CNN) were significantly higher than those of the indicated method at P < 0.05. ** P < 0.01. *** P < 0.001. RO radiomics, SVM support vector machine, RF random forest, LR logistic regression, MT mixed training, AT assistance training. Detailed performance values are provided in Supplementary Table 2 for the internal cross-validation cohort and in Supplementary Table 3 for the hold-out testing cohort.
