Published October 26, 2020
| Version v3
Software
Open
Pretrained nnU-Net Model from the cMRI M&Ms Challenge 2020
Authors/Creators
- 1. Division of Medical Image Computing, German Cancer Research Center (DKFZ); Heidelberg University, Medical Faculty Heidelberg, Heidelberg, Germany)
- 2. Division of Medical Image Computing, German Cancer Research Center (DKFZ)
Description
Pretrained model of the winning contribution of the "Multi-Centre, Multi-Vendor & Multi-Disease Cardiac Image Segmentation Challenge (M&Ms)".
The model is build with the nnU-Net framework and comprises an ensemble of five 2d and five 3d models.
The model can be downloaded (after setting up the nnU-Net framework) using following command
nnUNet_download_pretrained_model Task114_heart_mnms
Data needs to be available as 3D .nii.gz files. For further specification how to prepare your data see or follow this nnU-Net example on prostate MRI data
Once your data is prepared you can run inference with
# run prediction with 2d nnU-Net
nnUNet_predict -i <path_to_input_folder> -o <path_to_temporary_output_folder_2d> --save_npz -t 114 -m 2d -tr nnUNetTrainerV2_MMS
# run prediction with 3d nnU-Net
nnUNet_predict -i <path_to_input_folder> -o <path_to_temporary_output_folder_3d> --save_npz -t 114 -m 3d_fullres -tr nnUNetTrainerV2_MMS
# ensemble 2d and 3d predictions
nnUNet_ensemble -f <path_to_temporary_output_folder_2d> <path_to_temporary_output_folder_3d> -o <path_to_final_predictions>
Files
Task114_heart_MNMs.zip
Additional details
References
- Fabian Isensee, Paul F. Jäger, Simon A. A. Kohl, Jens Petersen, Klaus H. Maier-Hein "Automated Design of Deep Learning Methods for Biomedical Image Segmentation" arXiv preprint arXiv:1904.08128 (2020).
- Campello, Víctor M. et al.: Multi-Centre, Multi-Vendor & Multi-Disease Cardiac Image Segmentation. In preparation.