3D Medical Image Segmentation using Parallel Transformers

This is the official pytorch implementation of the TransHRNet

Requirements

CUDA 11.0

Python 3.7

Pytorch 1.7

Torchvision 0.8.2

Usage

1. Data Preparation

• Download BCV dataset (https://www.synapse.org/#!Synapse:syn3193805/wiki/217789)

• Preprocess the BCV dataset use the nnUNet package.

  cd nnUNet/nnunet/dataset_conversion 
  Run python Task017_BeyondCranialVaultAbdominalOrganSegmentation.py
  

• Training and Testing ID are in data/splits_final.pkl

2. Training

cd TransHRNet_package/TransHRNet/run

• Run python run_training.py -gpu="0" -outpath="TransHRNet_result" for training.

3. Testing

cd TransHRNet_package/TransHRNet/run

• Run python run_training.py -gpu='0' -outpath="TransHRNet_result" -val --val_folder='validation_result' for validation.

4. Performance on Multi-Atlas Labeling Beyond the Cranial Vault (BCV) dataset

Acknowledge

Part of codes are reused from the CoTr (https://github.com/YtongXie/CoTr). Thanks to Fabian Isensee for the codes of CoTr.