This is the official code for the paper "SPGNet: A Shape-Prior Guided Network for Medical Image Segmentation" (https://www.ijcai.org/proceedings/2024/0140.pdf).
Our paper has been accepted to IJCAI 2024.
You should download the datasets mentioned in the paper from the following links:
(1) BUSI: Breast Ultrasound Images Dataset
(2) miniJSRT(Segmentation02): Japanese Society of Radiological Technology Database
If necessary, run the code in ./dataset/split_masks.py to pre-separate the connected components.
Next, run the code in ./dataset/LMK_dataset.py to generate the landmarks data. You can also adjust the number of shape points in this file.
cd dataset
python LMK_dataset.py
pip install -r requirements.txt
You can set the hyperparameters in train.py according to your preferences.
python train.py
@inproceedings{song2024spgnet,
title={SPGNet: a shape-prior guided network for medical image segmentation},
author={Song, Zhengxuan and Liu, Xun and Zhang, Wenhao and Gong, Yongyi and Hao, Tianyong and Zeng, Kun},
booktitle={Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence},
pages={1263--1271},
year={2024}
}