• pytorch>=1.0
• tensorboardX
• scikit-image
• scipy
• tqdm

Note: the code has been tested on ubuntu. I'm not sure whether it works on windows.

• LA Heart MRI dataset: run python train_LA.py
• Liver tumor CT dataset: run python train_LITS.py

• LA Heart MRI dataset: run python train_LA_BD.py
• Liver tumor CT dataset: run python train_LITS_BD.py

You need to set --exp properly. Both compute_sdf and compute_sdf1_1 are worth to try.

• LA Heart MRI dataset: run python train_LA_HD.py
• Liver tumor CT dataset: run python train_LITS_HD.py

You need to set --exp properly. Both compute_dtm and compute_dtm01 are worth to try.

• LA heart MRI dataset: run python test_LA.py
• Liver tumor CT dataset: run python test_LITS.py

Xue et al. Shape-Aware Organ Segmentation by Predicting Signed Distance Maps arxiv

• run python train_LA_AAAISDF.py
• run python train_LA_AAAISDF_L1.py

• run test_LA_AAAISDF.py

Wang et al. Deep Distance Transform for Tubular Structure Segmentation in CT Scans arxiv

Navarro et al. Shape-Aware Complementary-Task Learning for Multi-organ Segmentation arxiv

• run python train_LA_MultiHead_FGDTM_L1.py to regress foreground distance transform map

L1 can be replaced with L2 or L1PlusL2

• run python train_LA_MultiHead_SDF_L1.py to regress signed distance function

L1 can be replaced with L2 or L1PlusL2

• run test_LA_MultiHead_FGDTM.py
• run test_LA_MultiHead_SDF.py

• run python train_LA_Rec_FGDTM_L1.py to regress foreground distance transform map

L1 can be replaced with L2 or L1PlusL2

• run python train_LA_Rec_SDF_L1.py to regress signed distance function

L1 can be replaced with L2 or L1PlusL2

• run test_LA_Rec_FGDTM.py
• run test_LA_Rec_SDF.py

• --model can be used to specificy the model name
• --epoch_num can be used to specificy the checkpoint