liver_spleen_segmentation

Liver and spleen segmentation using modified version of 2.5D DeepLabV3+

Train from scracth

1. Modify root_path, project_path, raw_data_path, result_path from config.py.
2. Place a shape file at raw_data_path/segment/trn_shp.npy. This .npy file should contain size of z-axis of each image since the input data will be flattened.
3. Place a data file at raw_data_path/segment/trn_dat.npy. This .npy file should contain raw dicom pixel values in the shape of (N, 512,512).
4. Place a segmentation mask file at raw_data_path/segment/trn_lbl.npy. This .npy file should contain integer values in the shape of (N, 512,512), (0-background, 1-spleen, 2-liver).
5. Place a segmentation mask file at raw_data_path/trn_bound_mask_2d.npy. This .npy file should contain boolean values of the liver and spleen boundary in the shape of (N, 512, 512).
6. run python main.py.

Run pretrained model

1. Download and unzip pretrained model.
2. Place .dcm files in a folder.
3. Modify ckpt and in_dir in test_seg.py.
4. run python test_seg.py