/dist-bias

Primary LanguageJupyter Notebook

Distribution Matching Losses Can Hallucinate Features in Medical Image Translation

aka How to Cure Cancer (in images) with Unpaired Image Translation

Joseph Paul Cohen, Margaux Luck, Sina Honari

https://arxiv.org/abs/1805.08841

Published at Medical Image Computing & Computer Assisted Intervention (MICCAI 2018). An abstract is published at the Medical Imaging with Deep Learning Conference (MIDL 2018)

How to run

Prepare the data

prepare_data.ipynb

Run the cyclegan for each split

$cd cyclegan
$sh run.sh

Requirements

If you want to use Conda:

conda create -n pytorch python=3 numpy scipy pandas scikit-learn
source activate pytorch
conda install pytorch torchvision cuda80 -c soumith

T-NT Dataset

If you are looking for the dataset used in this paper we have created a dataset called T-NT which contains MRI slides with and without tumors.

Download it here: https://academictorrents.com/details/d52ccc21455c7a82fd6e58964c89b7da99e0edf7

It includes segmentations:

Sample Flair Images

Tumor NoTumor