Figure 2.

Quantitative evaluation of the ${\mathrm{MGMT}}_{\mathrm{infr},\mathrm{all}}$ and ${\mathrm{MGMT}}_{\mathrm{pred},\mathrm{all}}$ predictions. (a) Violin plots of the distance measures of the probabilistic predictions show that the ${\mathrm{MGMT}}_{\mathrm{pred},\mathrm{all}}$ predictions are closer (statistically significant) to the target labels for test data. (b, c) ROC curves and (d, e) confusion matrices of ${\mathrm{MGMT}}_{\mathrm{infr},\mathrm{all}}$ and ${\mathrm{MGMT}}_{\mathrm{pred},\mathrm{all}}$ respectively along with cell-wise diagnosis breakdown. Note that ${\mathrm{MGMT}}_{\mathrm{infr},\mathrm{all}}$ has a tendency to over-excise lesions.