Summary

Resources to create a container that runs nnUNet model inference on AIMI collection out of the box

Running instructions

• Create an input_dir_ct containing list of dcm files corresponding to a given series_id for CT modality

• Create an input_dir_pt containing list of dcm files corresponding to a given series_id for PT modality

• Create an output_dir to store the output from the model. This is a shared directory mounted on container at run-time. Please assign write permissions to this dir for container to be able to write data

• Next, pull the image from dockerhub:

  • docker pull bamfhealth/bamf_nnunet_pt_ct_breast:latest

• Finally, let's run the container:

  • docker run --gpus all -v {input_dir_ct}:/app/data/input_data/ct:ro -v {input_dir_pt}:/app/data/input_data/pt -v {output_dir}:/app/data/output_data bamfhealth/bamf_nnunet_pt_ct_breast:latest

• Once the job is finished, the output inference mask(s) would be available in the {output_dir} folder

• Expected output from after successful container run is:

  • {output_dir}/seg_ensemble_primary.dcm -- if nifti to dcm conversion is a success
  • {output_dir}/seg_lesions_ensemble.nii.gz -- if nifti to dcm conversion is a failure

Breast PET/CT Segmentation

We used our FDG-PET/CT lesion segmentation model trained on AutoPET 2023 dataset to segment lesions for each of the five folds and ensemble the lesion segments. We then used segmentations from TotalSegmentator to limit the predicted lesions to only those in the breast area.

The model_performance notebook contains the code to evaluate the model performance on QIN-Breast collection against a validation evaluated by a radiologist and a non-expert.

Running the model

#TODO

Build container from pretrained weights

#TODO

Running inference

By default the container takes an input directory that contains DICOM files of PET/CT scans from the QIN-Breast collections, and an output directory where DICOM-SEG files will be placed. To run on multiple scans, place DICOM files for each scan in a separate folder within the input directory. The output directory will have a folder for each input scan, with the DICOM-SEG file inside.

example:

#TODO

There is an optional --nifti flag that will take nifti files as input and output.

Run inference on IDC Collections

This model was run on PET/CT scans from the QIN-Breast collection. The AI segmentations and corrections by a radioloist for 10% of the dataset are available in the breast-fdg-pet-ct.zip file on the zenodo record

You can reproduce the results with the run_on_idc_data notebook on google colab.

Training your own weights

Refer to the training instructions for more details. #TODO