Deformable-Medical-Image-Registration-via-Multiveiw-Adversarial-Learning

Implement of "Deformable-Medical-Image-Registration-via-Multiveiw-Adversarial-Learning"

Registration Network Architecture

Multiview Adversarial Learning

Prerequisites

• python==3.8.8
  
• pytorch==1.8.1

Installation

The required packages are located in requirements.

pip install pytorch==1.8.1 torchvision==0.9.1 torchaudio==0.8.1 cudatoolkit=10.2 -c pytorch
pip install -r requirement.txt

Dataset

Abdominal CT

BTCV Datasets Link : https://www.synapse.org/Synapse:syn3193805/wiki/217789
Abdominal 1K Datasets Link : https://github.com/JunMa11/AbdomenCT-1K

Organs consist of Spleen, Kidney, Gallbladder, Esophagus, Liver, Stomach, Aorta, Inferior Vena Cava, Pancreas, Right and Left Adrenal Gland.

Brain MRI

LPBA40 Link: https://www.loni.usc.edu/research/atlas_downloads

Brain MRI contains 40 T1-weighted MR images annotated with 56 subcortical ROIs. We combine the 56 labels into 7 region labels (i.e., Frontal Lobe, Parietal Lobe, Occipital Lobe, Temporal Lobe, Cingulate Lobe, Putamen, and Hippocampus) defined according to the main clinical structures of the brain.

Training

• Before training, pre-processing and initial affine registration must be performed.
  • For pre-processing, reference preprocessing_BTCV_1K.py, preprocessing_lpba40.py.
  • For initial affine registration, reference Train_Affine.py.
    

python Train_M_Adv.py \
    --affine_model experiments/affine \
    --dataset_dir Dataset/BTCV_Abdominal_1k \
    --save_validation_img True \
    --max_epoch 100 \

Inference

python Inference_M_Adv.py --affine_model experiments/affine --dataset_dir Dataset/BTCV_Abdominal_1k

Main results

Abdominal CT

• 3D Visualization
  

• 2D Visualization
  

Brain MRI

• 2D Visualization
  

Ablation Studied

Reference Codes

• https://github.com/junyuchen245/TransMorph_Transformer_for_Medical_Image_Registration
• https://github.com/voxelmorph/voxelmorph
• https://github.com/zju3dv/LoFTR
• https://github.com/microsoft/CvT