Moving from 2D to 3D: volumetric medical image classification for rectal cancer staging

Topics:

1. 2D vs. 3D vs. Mixtures of 2D and 3D
2. Supplementary loss function
3. Depth aggregation function

Common parser arguments:

1. --fusion
   → fr2d, fr3d, f2plus1d, fmc2, fmc3, fmc4, fmc5, frmc2, frmc3, frmc4, frmc5

2. --folder-name
   → folder name to save results

3. --node
   → User must register his/her computer information to config.ini

1. 2D vs. 3D vs. Mixtures of 2D and 3D

main_backbone_fw.py

2. Supplementary loss function

+ triplet loss

main_triplet_fw.py

+ center loss

main_center_loss_sgd_fw.py

3. 2D vs. 3D vs. Mixtures of 2D and 3D

main_vw.py

• Select frame aggregation function by --aggregation-function
  → bilinear, gap, mxp, attention