This is the repository for the paper "DeepAtlas: Joint Semi-Supervised Learning of Image Registration and Segmentation " at MICCAI 2019 [arxiv] by Zhenlin Xu and Marc Niethammer.
Install torch>=1.0
and torchvision
according to your config
Install other dependencies with pip install -r requirements.txt
The OAI knee MRIs and MindBoggle101 brain MRIs were used in the paper. We can only share our processed brain MRIs due to copyright limitations. Please download the preprocessed Mindboggle101 from Google Drive. Note that, in DeepAtlas paper, a few images were disregarded due the segmentation labeling errors that were fixed in the later versions of MB101.
E.g. Train a segmentation model with 21 training samples
python train_seg.py --num-samples 21 --data-root $DATA_ROOT --num-epochs 100 --lr 1e-3 --log-root ./logs
TODO
TODO