Skip to content

Latest commit

 

History

History
45 lines (32 loc) · 2.33 KB

README.md

File metadata and controls

45 lines (32 loc) · 2.33 KB

Image-to-image translation for biomedical image synthesis

This Code implements a variety of image-to-image translation models. These have been used in [1] to generate synthetic biomedical images for lesion segmentation in DWI images. Moreover, we provide a GAN implementation that can be used to synthesize stroke lesions. If you use this code, please cite our work [1].

Implemented models

  • cycleGAN [2]
  • pix2pix [3]
  • SPADE [4]
  • regular GAN

The models can be customized with different architectures (ResNet or UNet) and parameters (2D/3D) as specified in the settings file. The model's data generator accepts dicom, nifti and numpy arrays as input.

Usage

To use this code, you have to set up a python enviornment:

$ conda create -n yourenv python=3.7
$ conda activate yourenv
$ pip install -r requirements.txt

To train a model

$ python train.py -g [GPU identifier; None for CPU] \
                  -c [/path/to/settings_file.cfg] \
                  -o [output directory; inferred from settings file if ''] \
                  -t [dropout sampling 1/0; default 0] \
                  -m [multi-thread data generator 1/0; default 0] \
                  -p [path/to/pre-trained/model] \
                  -l [int identifier for pre-trained model]

The trained model will be saved in the output directory specified as an optional argument or in the savmodpath defined in the settings file. There, you will also find model logs and outputs.

An example setting is provided in setting/settings_example.cfg.

References

[1] Platscher M., Zopes J., & Federau C. (2020). Image Translation for Medical Image Generation--Ischemic Stroke Lesions. arXiv preprint arXiv:2010.02745.

[2] Zhu J. Y., Park T., Isola P., & Efros A. A. (2017). Unpaired image-to-image translation using cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer vision (pp. 2223-2232).

[3] Isola P., Zhu J. Y., Zhou T., & Efros A. A. (2017). Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 1125-1134).

[4] Park T., Liu M. Y., Wang T. C., & Zhu J. Y. (2019). Semantic image synthesis with spatially-adaptive normalization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 2337-2346).