Inherent Consistent Learning for Accurate Semi-supervised Medical Image Segmentation

The official repo for [MIDL'23 Oral] "Inherent Consistent Learning for Accurate Semi-supervised Medical Image Segmentation"

Introduction

We propose a novel Inherent Consistent Learning (ICL) method, which aims to learn robust semantic category representations through the semantic consistency guidance of labeled and unlabeled data to help segmentation.

Methods

Environment Setup

Installation

1. Clone the repo

git clone https://github.com/zhuye98/ICL.git 
cd ICL

2. Install torch and torchvision required packages.

Some important required packages include:

• torch == 1.9.1+cu111
• python == 3.7
• SimpleITK == 2.2.0
• monai == 1.0.1
• tensorboardX, numpy, h5py and more, please refer to requirements.txt

Data Preparation

Download the processed data and put the data in ../data/BraTS2019 or ../data/ACDC, please read and follow the README.

Run

Training on ACDC datasets:

cd code
# For 2D experiments (unet-based)
python train_inherent_consistent_unet_2D.py --root_path ..data/ACDC --exp ACDC/Unet_ICL --num_classes 4 --labeled_num 3/7

# For 2D experiments (swinunet-based)
python train_inherent_consistent_swinunet_2D.py --root_path ..data/ACDC --exp ACDC/Swin_ICL --num_classes 4 --labeled_num 3/7

Training on BraTS datasets:

# For 3D experiments on BraTS (3d unet-based)
python train_inherent_consistent_unet_3D_BraTS.py --root_path ..data/BraTS19 --exp BraTS19/Unet_ICL --num_classes 2 --labeled_num 25/50 --use_ssl_pretrained

# For 3D experiments on BraTS (3d swinunetr-based)
python train_inherent_consistent_swinunetr_3D_BraTS.py --root_path ..data/BraTS19 --exp BraTS19/Unet_ICL --num_classes 2 --labeled_num 25/50 --use_ssl_pretrained

Training on AMOS datasets:

# For 3D experiments on AMOS (3d unet-based)
python train_inherent_consistent_unet_3D_AMOS22.py --root_path ..data/AMOS --exp AMOS/Unet_ICL --num_classes 16 --labeled_num 15 --val_num 30

Training on different datasets:

python test_2D_ACDC.py / test_3D_AMOS.py / test_3D_BraTS

Acknowledgements

Our code is origin from SSL4MIS. We are grateful to these authors for their valuable contributions, and I am hopeful that our newly proposed method can also contribute to advancing related Semi-supervised Learning research.