UTSRMorph: A Unified Transformer and Superresolution Network for Unsupervised Medical Image Registration. (TMI2024)
September 26, 2024 ยท View on GitHub
Keywords: Deformable image registration, ConvNets, Transformer, Cross-attention, Superresolution.
Here is the PyTorch implementation of the paper:
Update progress
23/9/2024 - The paper is accepted in IEEE TMI.
31/8/2024 - UTSRMorph trained in Abdominal MR-CT and CMF tumor MR-CT datasets is now publicly available!
4/24/2024 - UTSRMorph trained in OASIS datasets with dice loss is improved and the model trained in IXI datasets is publicy available!
4/15/2024 - UTSRMorph trained in OASIS datasets is now publicly available!
Requirments
We trained our models depending on Pytorch 1.13.1 and Python 3.8.
Train and infer
UTSRMorph are tested on 4 datasets: OASIS, IXI, Abdominal MR-CT and CMF tumor MR-CT datasets.
If you want to train OASIS dataset, you only need to run the following script: train_UTSRMorph_oasis.py. After the training stage, the model will be saved in experients folder.
To infer the trained model, you just need to run infer_UTSRMorph.py script.
The rest 3 datasets are the same as OASIS, the only difference is the path of dataset.
Datasets
4 datasets: OASIS, IXI, Abdominal MR-CT and CMF tumor MR-CT dataset. The IXI and OASIS dataset can be downloaded from TransMorph. You can download the Abdominial MR-CT dataset from Abdominial MR-CT, the afterprocessed dataset can be downloaded from Abdominial MR-CT. The CMF tumor MR-CT dataset is avaiable on Google Drive.
Contact
If you have any questions, feel free to contact zhangrunshi@buaa.edu.cn