Deep Active Contour Network for Medical Image Segmentation
- Tensorflow 1.x
- Put '.nii' format data into directory as the following file structure:
Dataset
└── training_data
├── training_data_nii
| ├── slices
│ │ └── CC0001_philips_15_55_M.nii.gz...
│ └── masks
│ └── CC0001_philips_15_55_M_ss.nii.gz...
│
├── valid_data_nii
| ├── slices
│ │ └── CC0016_philips_15_55_M.nii.gz...
│ └── masks
│ └── CC0016_philips_15_55_M_ss.nii.gz...
│
├── test_data_nii
| ├── slices
│ │ └── 5254HD.nii.gz...
│ └── masks
│ └── 5254HD.manual.mask.nii.gz...
...
- run
input.mto slice the volumn into 2d image. - run
create_h5.pyto convert dataset into hdf5 file. - train：
python main.py
valid：python main.py --action=test
predict：python main.py --action=predict - run
save_pred_and_quantify.mto convert the prediction results into different folders according to different volume. - run
save_results.mto convert 2d results into 3d volumes.
- test on brainsuite data:
-
Dense Unet Structure
-
DACN Structure
- Training data : CC0001 - CC0015, CC0110-CC0134, CC0230-CC0254, CC0350-CC0359
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent