Skip to main content
. 2021 Feb 13;13(4):786. doi: 10.3390/cancers13040786

Figure A1.

Figure A1

Learning curves for the different models. Values are averaged over the ten times each model was trained; shadowed area represents the standard deviation. The C3D pre-trained model performed best with no overfitting. The 3D model trained from scratch and the ImageNet pre-trained 2D model started to overfit early, at around epoch 40 and 15, respectively. The relative large difference between training and validation performance stems from application of an external validation set.