Install-nnUnet.md

Install nnUnet

• out-of-the-box segmentation algorithm pip install nnunetv2
• to modify code, open a terminal, select a directory:

git clone https://github.com/MIC-DKFZ/nnUNet.git
cd nnUNet
pip install -e .

1. Create 3 folders in folder nnUNet, nnUNet_raw, nnUNet_preprocessed, nnUNet_results
2. Set environment variables
   • Linux, temporary in terminal (permanent in home folder .bashrc):
   • Windows, temporary in Command prompt (permanent in Edit Environment Variable GUI):
   • in code: /home/jing/pycode/nnUNet/nnunetv2/paths.py

# in Linux
export nnUNet_raw='/home/jing/pycode/nnUNet/nnUNet_raw'
export nnUNet_preprocessed='/home/jing/pycode/nnUNet/nnUNet_preprocessed'
export nnUNet_results='/home/jing/pycode/nnUNet/nnUNet_results'
# in Windows
set nnUNet_raw=C:/Users/jing/pycode/nnUNet_raw
set nnUNet_preprocessed=C:/Users/jing/pycode/nnUNet_preprocessed
set nnUNet_results=C:/Users/jing/pycode/nnUNet_results
# in code, added by user
os.environ['nnUNet_raw'] = r'/home/jing/pycode/nnUNet/nnUNet_raw'
os.environ['nnUNet_preprocessed'] = r'/home/jing/pycode/nnUNet/nnUNet_preprocessed'
os.environ['nnUNet_results'] = r'/home/jing/pycode/nnUNet/nnUNet_results'

3. Dataset conversion
   in file /home/jing/pycode/nnUNet/pyproject.toml, a sentence like

nnUNetv2_convert_MSD_dataset = "nnunetv2.dataset_conversion.convert_MSD_dataset:entry_point"

shows the path, along the path find out file convert_MSD_dataset.py, in the last line, change the parameter, then run it.

convert_msd_dataset('/home/jing/pycode/nnUNet/nnUNet_raw/Task05_Prostate', overwrite_target_id=1101)

4. Experiment planning and preprocessing
   • in terminal nnUNetv2_plan_and_preprocess -d DATASET_ID --verify_dataset_integrity # DATASET_ID is 1101
   • in code, search nnUNetv2_plan_and_preprocess in the file pyproject.toml, you will find:

nnUNetv2_plan_and_preprocess = "nnunetv2.experiment_planning.plan_and_preprocess_entrypoints:plan_and_preprocess_entry"

then go to this file plan_and_preprocess_entry.py, then run it after adding a default parameter in function plan_and_preprocess_entry:

parser.add_argument('-d', nargs='+', type=int,
help="[REQUIRED] List of dataset IDs. Example: 2 4 5. This will run fingerprint extraction, experiment "
"planning and preprocessing for these datasets. Can of course also be just one dataset", default=[1101])

5. Model training
   • in terminal nnUNetv2_train DATASET_NAME_OR_ID UNET_CONFIGURATION FOLD [additional options, see -h]
   • in code, search nnUNetv2_train in the file pyproject.toml, you will find:

nnUNetv2_train = "nnunetv2.run.run_training:run_training_entry"

then go to this file run_training_entry.py, then run it after adding a default parameter in function run_training_entry:

parser.add_argument('-dataset_name_or_id', type=str, default='1101', required=False, help="Dataset name or ID to train with")
parser.add_argument('-configuration', type=str, required=False, default='2d', # or '3d_fullres'
help="Configuration that should be trained")
parser.add_argument('-fold', type=str, default='1', required=False, help='Fold of the 5-fold cross-validation. Should be an int between 0 and 4.')

modify training epochs (default is 1000) before running the code:
/home/jing/python_code/nnUNet/nnunetv2/training/nnUNetTrainer/nnUNetTrainer.py ctrl+f: self.num_epochs = 2

6. Best configuration (optional)
   nnUNet_find_best_configuration -m 2d 3d_fullres 3d_lowres 3d_cascade_fullres -t 1

7. Model predicting

   • in terminal nnUNet_predict -i .../Task05_Prostate/imagesTs/ -o .../Task05_Prostate/inferTs/ -t 1 -m 3d_fullres -f 0

   • in code, search nnUNetv2_predict in the file pyproject.toml, you will find:

nnUNetv2_predict = "nnunetv2.inference.predict_from_raw_data:predict_entry_point"

8. Model evaluating
   • in terminal nnUNet_evaluate_folder -ref LABELFOLDER -pred PREDICTIONFOLDER -l 1 2 3
   • in code, search nnUNetv2_evaluate_folder in the file pyproject.toml, you will find:

nnUNetv2_evaluate_folder = "nnunetv2.evaluation.valuate_predictions:evaluate_folder_entry_point"

✴️ reference: https://github.com/MIC-DKFZ/nnUNet/blob/master/documentation/how_to_use_nnunet.md