metadata
license: cc-by-4.0
This repository provides the nnUNet v2 checkpoint for multi-organ lesion segmentation from abdominal CT images.
The model was trained on a large-scale multi-center CT dataset with expert-annotated lesion masks across multiple organs (liver, pancreas, kidney, colon).
Label Definitions
"labels": {
"background": 0,
"liver_segment_1": 1,
"liver_segment_2": 2,
"liver_segment_3": 3,
"liver_segment_4": 4,
"liver_segment_5": 5,
"liver_segment_6": 6,
"liver_segment_7": 7,
"liver_segment_8": 8,
"pancreas_head": 9,
"pancreas_body": 10,
"pancreas_tail": 11,
"kidney_left": 12,
"kidney_right": 13,
"colon": 14,
"liver_lesion": 15,
"pancreatic_lesion": 16,
"kidney_lesion": 17,
"colon_lesion": 18
}
1. Installation
git clone https://github.com/MIC-DKFZ/nnUNet.git
cd nnUNet
pip install -e .
2. Inference
2.1 Create a folder (e.g., nnUNet_eval), rename all your ct files and put them in the folder.
nnUNet_eval/
└── Dataset1351/
└── imagesTs/
├── ct001_0000.nii.gz
├── ct002_0000.nii.gz
2.2 Place the checkpoint into
nnUNet_results/
└── Dataset1351/nnUNetTrainer__nnUNetResEncUNetLPlans__3d_fullres/fold_all/
2.3 Run inference
export nnUNet_N_proc_DA=36
export nnUNet_results="./nnUNet_results"
export nnUNet_predictions="./nnUNet_predictions"
export nnUNet_eval="./nnUNet_eval"
GPU_ID=0
DATASET=1351
TRAINER=nnUNetTrainer
CUDA_VISIBLE_DEVICES=$GPU_ID nnUNetv2_predict -d $DATASET -i $nnUNet_eval/ -o $nnUNet_predictions/ -tr $TRAINER -d $DATASET -c 3d_fullres -f all -p nnUNetResEncUNetLPlans --continue_prediction