Unsupervised Deformable Image Registration with Absent Correspondences in Pre-operative and Post-Recurrence Brain Tumor MRI Scans

This is the official Pytorch implementation of "Unsupervised Deformable Image Registration with Absent Correspondences in Pre-operative and Post-Recurrence Brain Tumor MRI Scans" (MICCAI 2022), written by Tony C. W. Mok and Albert C. S. Chung.

Prerequisites

• Python 3.5.2+
• Pytorch 1.3.0 - 1.9.1
• NumPy
• NiBabel
• Scipy

This code has been tested with Pytorch 1.10.0 and NVIDIA TITAN RTX GPU.

Inference

Inference for DIRAC:

python BRATS_test_DIRAC.py

Inference for DIRAC-D:

python BRATS_test_DIRAC_D.py

Train your own model

Step 1: Download the BraTS-Reg dataset from https://www.med.upenn.edu/cbica/brats-reg-challenge/

Step 2: Define and split the dataset into training and validation set, i.e., 'Dataset/BraTSReg_self_train' and 'Dataset/BraTSReg_self_valid', respectively.

Step 3: python BRATS_train_DIRAC.py to train the DIRAC model or python BRATS_train_DIRAC_D.py to train the DIRAC-D model.

Publication

If you find this repository useful, please cite:

• Unsupervised Deformable Image Registration with Absent Correspondences in Pre-operative and Post-Recurrence Brain Tumor MRI Scans
  Tony C. W. Mok, Albert C. S. Chung
  MICCAI 2022. eprint arXiv:2206.03900

• Conditional Deformable Image Registration with Convolutional Neural Network
  Tony C. W. Mok, Albert C. S. Chung
  MICCAI 2021. eprint arXiv:2106.12673

Keywords

Keywords: Absent correspondences, Patient-specific registration, Deformable registration