This is a PyTorch implementation based on the MICCAI paper by Qian Yue et al. on LGE-CMR segmentation. UNet, SRNN, SCN and SRSCN are performed for the segmentation task.
Automatic cardiac segmentation from LGE-CMR is of great clinical value. In [3], Yue et al. proposed SRSCN, a U-Net based method incorporating additional modules for shape reconstruction and spatial constraint. The pipeline from their paper summarizes the model. For more details, please refer to [3].
In this project, we trained the basic UNet, SRNN, SCN and SRSCN on MSCMR dataset, which is available upon registration.
python main.py --path "data_path" --batch_size 8 --dim 240 --lr 1e-4 --threshold 0.65 --end_epoch 30
The data path is organized as follows:
data/
-- image files & gt files
-- train.txt (with each line: image_path gd_path z_index)
-- validation.txt (with each line: image_path gd_path z_index)
-- test.txt (with each line: image_path dx dy dz)
Pretrained models with bathsize = 8 and epoch = 30 are stored in checkpoints/model_name.
To use the models for segmentation, please prepare test.txt in the data path as described above and type:
python predict.py --load_path checkpoints/"model name" --predict_mode multiple --threshold 0.6 --dim 240
[1]Xiahai Zhuang: Multivariate mixture model for myocardial segmentation combining multi-source images. IEEE Transactions on Pattern Analysis and Machine Intelligence 41(12), 2933–2946, 2019
[2]Xiahai Zhuang: Multivariate mixture model for cardiac segmentation from multi-sequence MRI. MICCAI 2016, 581–588, Springer, 2016
[3]Q Yue, X Luo, Q Ye, L, Xu, X Zhuang. Cardiac Segmentation from LGE MRI Using Deep Neural Network Incorporating Shape and Spatial Priors. MICCAI 2019, LNCS 11765, pp. 559-567, 2019. https://github.com/xzluo97/LGE_SRSCN