Skip to main content
. 2022 Nov 10;12:19200. doi: 10.1038/s41598-022-21848-3

Figure 8.

Figure 8

ROC curves for each of the seven chosen comparative CNN models (training set from 80% of FBCG dataset used previously for the fivefold stratified CV) on test set (20% of FBCG dataset). It shows that, on average, all the chosen models exhibited highest performance in identifying Grade 0 and lowest performance in identifying Grade 1 (except MobileNetV2).