igip-liu / SLC-Net

SLC-Net

This repository is for our paper "Semi-supervised medical image segmentation using cross-style consistency with shape-aware and local context constraints"

Requirements

Some important required packages include:

• Pytorch version >=0.4.1.

• TensorBoardX

• Python == 3.7

• Efficientnet-Pytorch

• Some basic python packages such as Numpy, Scikit-image, SimpleITK, Scipy，Batchgenerators ......

Usage

1、Clone the repo;

git clone https://github.com/igip-liu/SLC-Net.git

2、Data Preparation;

The division method of training/validation/test set can be seen:

ACDC dataset

Prostate dataset

NIH Pancrea dataset

The data that can be used to train our code can be seen:

ACDC dataset

Prostate dataset

NIH Pancrea dataset

The division of labeled/unlabeled datasets can be found in this code

You can regenerate the training data:

cd SLC-Net/code/dataloaders

python acdc_data_processing.py

3、Train the model;

cd SLC-Net/code

CUDA_VISIBLE_DEVICES=3 python train_CLB.py --root_path ../data/ACDC --exp ACDC/SLC-Net --num_classes 4 --labeled_num 7 --use_block_dice_loss --block_num 4

4、Test the model;

cd SLC-Net/code

CUDA_VISIBLE_DEVICES=0 python test_2D_fully.py --root_path ../data/ACDC --exp ACDC/SLC-Net --num_classes 4 --labeled_num 7

Our code is based on the UAMT, SSL4MIS and Dual-Normalization. Thanks for these authors for their valuable works.