amrsegnet's Introduction

AMRSegNet:Adaptive Modality Recalibration Network for Lung Tumor Segmentation on Multi-modal MR Images

This is the Pytorch implementation of AMRSegNet for paper《Adaptive Modality Recalibration Network for Lung Tumor Segmentation on Multi-modal MR Images》.

Installation

Install pytorch with python 3.7, pytorch==1.4.0, torchvision==0.5.0, CUDA==10.1.
Python package requirement: SimpleITk, pydicom, tensorboardX
Clone this repository:

git clone https://github.com/Nicholasxin/AMRSegNet  
cd AMRSegNet

Dataset

Our T2W-DWI MR dataset is private. For code implementation, the dataset for training and testing consist of T2W slices, DWI slices, label slices, which are all paired.

Training

In the folder of repository AMRSegNet, open terminal and run python train.py.
Note: adding --ngpu to alter to the number of GPUs, adding --batchSz to change the batch size, adding --nEpochs to set the number of training epochs.
For showing the training process on tensorboard, the folder runs will be created. The trained model will be saved in auto-created folder work.
To open the tensorboard, open terminal and run tensorboard --logdir runs.

Testing

run python train.py with --inference following the path of inference T2W data, --dwiinference following the path of inference DWI data, --target following the path of label of T2W data, --resume following the path of the best saved training model. All the added commands are requisite.