Skip to content

Latest commit

 

History

History
89 lines (80 loc) · 2.58 KB

readme.md

File metadata and controls

89 lines (80 loc) · 2.58 KB
Raw

Discriminative Cross-Modal Data Augmentation for Medical Imaging Applications

This is the Pytorch implementaion of the paper:

Discriminative Cross-Modal Data Augmentation for Medical Imaging Applications

Dependencies

  • torchvision
  • torch
  • tensorboard
  • dominate
  • visdom
  • fastai
  • tensorboardX

File Orgnization

- generate_part    :Generate fake CT images from other modalities
    - mri2ct.sh    :Train translator (mri -> ct)
    - pet2ct.sh    :Train translator (prt -> ct)
    - mri_generate.sh   :Generate fake CT images from MRI images using trained translator
    - pet_generator.sh  :Generate fake CT images from PET images using trained translator

- task_part    :Combining fake and real images to do prediction
    - mr_ct.sh    :Do prediction using real and fake(mr2ct) CT images
    - pet_ct.sh   :Do prediction using real and fake(pet2ct) CT images

- datasets 
    - real_ct   :real CT images
    - real_mri  :real MRI images
    - real_pet  :real PET images
    - fake_mr2ct    :fake CT images (MRI->CT)
    - fake_pet2ct    :fake CT images (PET->CT)
    - mri
        - trainA    :Train set of source modality images (Here is MRI)
        - trainB    :Train set of target modality images (Here is CT)
        - valA    :Validation set of source modality images (Here is MRI)
        - valB    :Validation set of target modality images (Here is CT)
    - pet
        - trainA    :Train set of source modality images (Here is PET)
        - trainB    :Train set of target modality images (Here is CT)
        - valA    :Validation set of source modality images (Here is PET)
        - valB    :Validation set of target modality images (Here is CT)

Datasets

We use three datasets:

  1. MRI dataset: Link to MRI dataset
  2. PET dataset: Link to PET dataset
  3. CT dataset: Link to CT dataset

Usage

1. Download datasets to correponding directories.

2. Train translator.

For MRI -> CT:

cd generate_part
sh mri2ct.sh

For PET -> CT:

cd generate_part
sh pet2ct.sh

3. Using trained translator to generate fake CT images.

For MRI -> CT:

cd generate_part
sh mri_generate.sh

For PET -> CT:

cd generate_part
sh pet_generate.sh

4. Physiological age prediction using real and fake images.

For MRI + CT:

cd task_part
sh mri_ct.sh

For PET + CT:

cd task_part
sh pet_ct.sh

References

CycleGAN: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix