How to run inference on test images using your trained model?

Once your have trained some models using the training script described in trainddp.md, you have model(s) that could be used for predicting the segmentation masks for test images. Running inference primarily uses three files from this codebase: config.py, segmentation/initialize_train.py, and segmentation/inference.py. Ensure that the config.py is correctly initialized (as described in trainddp.md) so that the inference code can find the path to the test images.

Step 1: Activate the required conda environment (lymphoma_seg) and navigate to segmentation folder

First, activate the conda environment lymphoma_seg using (created as described in conda_env.md):

conda activate lymphoma_seg
cd segmentation

Step 2: Run the inference script

After this, run the following script in your terminal. Note: we run the inference only on one GPU (denoted by cuda:0 in your machine).

python inference.py --fold=0 --network-name='unet' --input-patch-size=192 --num_workers=2  --sw-bs=2 

• inference.py is the inference code that this script is using.

• --fold defines which fold's trained model you want to use for inference. When training script is run for the first time, two files, namely, train_filepaths.csv and test_filepaths.csv gets created within the folder WORKING_FOLDER/data_split, where the former contains the filepaths (CT, PT, mask) for training images (from imagesTr and labelsTr folders as described in dataset_format.md), and the latter contains the filepaths for test images (from imagesTs and labelsTs), respectively. The purpose of setting fold in this case is not to point to the specific fold dataset (since we are only using the test set for inference), but to define which fold's trained model to use. Defaults to 0.

• --network-name defines the name of the network. In this work, we have trained UNet, SegResNet, DynUNet and SwinUNETR (adpated from MONAI [LINK]). Hence, the --network-name should be set to one of unet, segresnet, dynunet, or swinunetr. Defaults to unet.

• --input-patch-size defines the size of the cubic input patch that is cropped from the input images during training. We used input-patch-size of 224 for UNet, 192 for SegResNet, 160 for DynUNet and 128 for SwinUNETR. Defaults to 192.

• --num-workers defines the num_workers argument inside training and validation DataLoaders. Defaults to 2.

• --sw-bs defines the batch size for performing the sliding-window inference via monai.inferers.sliding_window_inference on the test inputs. Defaults to 2.

Alternatively, modify the segmentation/predict.sh script for your use-case (which contains the same bash script as above) and run:

bash predict.sh

The predicted masks will be written to LYMPHOMA_SEGMENTATION_FOLDER/results/predictions/fold{fold}/{network_name}/{experiment_code}, as described in results_format.md file. The predicted masks are assigned the same filenames as the corresponding original ground truth segmentation masks. The relevant directory structure may then look like:

└───lymphoma.segmentation/
        ├── data
        └── results
            ├── logs
            ├── models
            └── predictions
                └── fold0
                    └── unet
                        └── unet_fold0_randcrop192
                            ├── Patient0003_20190402.nii.gz
                            ├── Patient0004_20160204.nii.gz 
                            ├── ...