[IEEE ISBI'26 Oral] Scribble-Supervised Medical Image Segmentation with Dynamic Teacher Switching and Hierarchical Consistency

Introduction

The overall framework of SDTNet:

@article{nguyen2026scribble,
  title={Scribble-Supervised Medical Image Segmentation with Dynamic Teacher Switching and Hierarchical Consistency},
  author={Nguyen, Thanh-Huy and Cao, Hoang-Loc and Chung, Dat T and Vu, Mai-Anh and Nguyen, Thanh-Minh and Le, Minh and Huynh, Phat K and Bagci, Ulas},
  journal={arXiv preprint arXiv:2601.14563},
  year={2026}
}

Datasets and Model Weights

1. ACDC Dataset

• Mask Annotations: ACDC
• Scribble annotations: ACDC scribbles

2. MSCMR Dataset

• Mask Annotations: MSCMRseg
• Scribble annotations: MSCMR_scribbles
• Scribble-annotated dataset for training: MSCMR_dataset.

We have organized the datasets and our model checkpoints, and they are now available for download at: 👉 Google Drive

Setup

1. Clone this project and create a conda environment

git clone https://github.com/loc110504/SDTNET.git
cd SDTNET

conda create -n env python=3.10
conda activate env

2. Install requirements and packages

pip install -r requirements.txt

Usage

1. For training:

cd code/train
python train_method_acdc.py # ACDC
python train_method_mscmr.py # MSCMRseg

2. For testing:

cd code/test
python test_acdc.py # ACDC
python test_mscmr.py # MSCMRseg

Acknowledgement

We acknowledge the public release of WSL4MIS, CycleMix and MAAG for their codes and processed dataset.