We propose a novel semi-supervised shadow aware network with boundary refinement (SABR-Net) to perform ultrasound images segmentation, which always have the challengs about the presence of shadow artifacts.
It is highly recommanded to adopt Conda/MiniConda to manage the environment to avoid some compilation errors.
- Clone the repository
git clone https://github.com/Swecamellia/semi-supervised-shadow-aware-network.git
- Install the dependencies
- Python 3.9
- Pytorch 1.11.0
- Cuda 11.3
- Other packages
pip install -r requirements.txt
We evaluated the proposed semi-supervised US segmentation method using two publicly accessible US image datasets. The first dataset, CAMUS [1], is a multi-structure cardiac dataset for US segmentation comprising clinical examinations of 500 patients. The second dataset, TN-SCUI [2], contains 3644 thyroid nodules of 3644 patients, manually annotated by experienced radiologists.
[1] S. Leclerc, E. Smistad, J. Pedrosa, A. Østvik, F. Cervenansky, F. Espinosa, T. Espeland, E. A. R. Berg, P.-M. Jodoin, T. Grenier, et al., “Deep learning for segmentation using an open large-scale dataset in 2d echocardiography,” IEEE Trans. Med. Imag., vol. 38, no. 9, pp. 2198–2210, 2019.
[2] H. Gireesha and S. Nanda, “Thyroid nodule segmentation and classification in ultrasound images,” Int. Journal of Engineering Research and Technology, 2014.
-
python prepare_inception.py:
First prepare inception features for own dataset.
-
python train_seg_pretrain.py：
Divide data of different proportions and mask the images, implement the inpainting and segmentation tasks of labeled and unlabeled data.
python train_seg_finetune.py:
To brige the gap between US images processed with shadow imitation operation and real US images.
python test.py
Results Obtained on Multi Structure Cardiac and Thyroid Ultrasound Datasets
