This is the official implementation of the paper:
Shumao Pang, Chunlan Pang, Zhihai Su, Liyan Lin, Lei Zhao, Yangfan Chen, Yujia Zhou, Hai Lu, Qianjin Feng. DGMSNet: Spine Segmentation for MR Image by a Detection-Guided Mixed-supervised Segmentation Network, Medical Image Analysis, 2022, 102261.
Website: https://www.researchgate.net/profile/Shumao_Pang2
- Pytorch 1.7.0
- Python 3.6.5
Here, we will show how to prepare the Dataset-B (i.e., Spinal disease dataset).
Step 1: Download the Spinal Disease Dataset from: https://tianchi.aliyun.com/dataset/dataDetail?dataId=79463.
Step 2: Move and rename the downloaded dataset to ./data/Spinal_disease_dataset. The files structure is shown as follows:
data
Spinal_disease_dataset
lumbar_train150
lumbar_train51
lumbar_testA50
lumbar_testB50
lumbar_train150_annotation.json
lumbar_train51_annotation.json
testA50_sagittal_image_information.xlsx
testB50_sagittal_image_information.xlsx
split_ind_fold1.npz
in
mask # Note that this is the manual segmentation annotation released by us for the lumbar_testA50 and lumbar_testB50.
keypoints
Step 3: Run the following commands in the terminal:
cd DGMSNet
python -u prepare_spinal_disease_dataset.py --data_dir=./data/Spinal_disease_dataset
run the following script in the terminal:
nohup ./train_spinal_disease_dataset.sh > train_spinal_disease_dataset.out &
To test the model without DGLF, please run the following script in the terminal:
nohup ./test_spinal_disease_dataset_without_dglf.sh > test_spinal_disease_dataset_without_dglf.out &
To test the model with DGLF, please run the following script in the terminal:
nohup ./test_spinal_disease_dataset_with_dglf.sh > test_spinal_disease_dataset_with_dglf.out &
The core and nn_tools packages were modified from https://github.com/wolaituodiban/spinal_detection_baseline.git and https://github.com/wolaituodiban/nn_tools.git respectively. Thanks for these works.
Shumao Pang, Chunlan Pang, Zhihai Su, Liyan Lin, Lei Zhao, Yangfan Chen, Yujia Zhou, Hai Lu, Qianjin Feng. DGMSNet: Spine Segmentation for MR Image by a Detection-Guided Mixed-supervised Segmentation Network, Medical Image Analysis, 2022, 102261.