Abdominal Multi-organ Segmentation via 3DUNet

Trained on LION data

Running Instructions

1. Clone the repository
2. Download the prerequisites from environment.yml
3. From Releases
   • Download 3dunet_model_2_channel.pth.tar to '/TrainedModels/3DUNet_2_Channel'
4. Copy the inference dataset into '/Custom_Datasets/<Your_Dataset_Name>'. This directory can have 2 filestructures

Structure 1

Your_Dataset_Name

Subjects_Dir

Subject_01

Subject_01_Fat_fused.nii.gz
Subject_01_Water_fused.nii.gz

...
Subject_ZZ

...

Structure 2

Your_Dataset_Name

Subjects_Dir

Subject_01

Subject_01_V1

Subject_01_V1_Fat_fused.nii.gz
Subject_01_V1_Water_fused.nii.gz

Subject_01_V2

...

...
Subject_ZZ

...

5. Ensure that for each subject, the keywords denoting Fat, Water, T2* images remain identical.
6. While using the appropriate args, run /Executables/predict_custom_input.py
   • --dataset_name: Enter the name of the dataset <Your_Dataset_Name>. By default, it is Dataset_Name
   • --fat_keyword: Enter the unique keyword for Fat maps. By default, it is Fat_fused
   • --water_keyword: Enter the unique keyword for Water maps. By default, it is Water_fused
   • --gpus GPU id. By default, it is set to 0
7. The images will be resized to (256, 224, 72) for the 3DUNet into '/Custom_Datasets/<Your_Dataset_Name>/Interpolated_Subjects_Dir', and the corresponding segmentations into '/Custom_Datasets/<Your_Dataset_Name>/Predicted_Masks'