GAN-enhanced Conditional Echocardiogram Generation

This repository accompanies our manuscript in the Medical Imaging Meets NeurIPS workshop of the 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada.
The draft of the manuscript is available here: https://arxiv.org/abs/1911.02121.

If you found this code useful in your research, please consider citing:

@article{abdi2019echoGen,
Author = {Amir H. Abdi and Teresa Tsang and Purang Abolmaesumi},
Title = {GAN-enhanced Conditional Echocardiogram Generation},
Year = {2019},
journal={arXiv preprint arXiv:1911.02121},    
}

Requirements &

• The implementation uses Keras with TensorFlow backend.
• scikit-image, SimpleITK, and matplotlib are used for data augmentation and visualization.
• Training is logged using the Weights & Biases tool (wandb).

Install the requirements by running

pip3 install -r requirements.txt

Alternatively, make a image by running:

docker build -t echo-generation .

A ready-to-use docker image is also available on docker hub.

Data

We use the publicly available dataset of CAMUS, which can be downloaded from here.

Training

To train the model call main.py with a config file of your choosing. The 5 config files corresponding to the 5 experiments of the article are available in the configs/ directory, e.g.

 python3 src/main.py \
 --dataset_path=$DATASETS/CAMUS \
 --config=configs/ventricle.json

The environment variable $DATASET is assumed to be set to where the CAMUS dataset directory is stored.

Sample Generated Echos

Check the complete video on YouTube.