Prostate-Segmenter

Prostate-Segmenter

‹ Go back to models

Model parameters

  • Task: Whole-gland prostate segmentation in pelvic Axial T2-W MRI scans
  • Imaging modality: MRI
  • Organ: Prostate
  • Input(s): Pelvic T2-W MRI
  • Output(s): Prostate gland binary segmentation
  • Training set size: 60 / Test set size: 30
  • Performance: Dice: %85
  • contact: Alireza Mehrtash, email: mehrtash at bwh.harvard.edu
  • Size: 6.8 GB

Demo

Command-line guide

Download Docker Image

docker pull deepinfer/deepinfer-prostate

Example

docker run -t -v ~/data/prostate_test/:/data deepinfer/prostate\
                   --ModelName prostate-segmenter\
                   --Domain BWH_WITHOUT_ERC\
                   --InputVolume /data/prostate.nrrd \
                   --OutputLabel /data/output_prostate_label.nrrd \
                   --ProcessingType Accurate\
                   --Inference Ensemble\
                   --verbose

Inputs

[Mandatory]
ModelName: (prostate-segmenter)
Domain: (BWH_WITH_ERC, BWH_WITHOUT_ERC, PROMISE12)
    Select the domain of trained models: 3 different domains are available:
    - BWH_WITH_ERC is a domain trained on pre-operative T2-Weighted images of Brigham and Women's Hosptial
    with endorctal coil on 3T MRI machine.
    - BWH_WITHOUT_ERC is a domain trained on pre-operative T2-Weighted images of Brigham and Women's Hosptial
    with endorctal coil on 3T MRI machine.
    - PROMISE12 are models that are trained on PROMISE12 challenge training dataset (multi-center multi-vendor dataset) 
    (https://promise12.grand-challenge.org/)
InputVolume: (an existing filename locating the T2-Weighted Pelvic MRI containing MRI)
OutputLabel: (output path of the prostate gland label)
ProcessingType: (Fast, Accurate)
    Accurate models use higher resolution inputs (0.27-0.625 mm in-plane resolutions) while fast models use
    1 mm in-plane resolutions.
Inference: (Single, Ensemble)
    Single: the prediction would be the output of a single model. Ensemble: the prediction will be the 
    calculated by ensembling of 5 models from 5-fold cross validation and majority voting.
[Optional]
verbose : 
    verbose mode for printing additional details about the procedure.

Reuse and Citations

The Prostate-Segmenter model is licensed under Slicer License.
For attribution in academic contexts, please cite the following work(s):

Mehrtash A. et al. "DeepInfer: open-source deep learning deployment toolkit for image-guided therapy." SPIE 2017.

BibTeX citation

@inproceedings{mehrtash2017deepinfer,
  title={DeepInfer: Open-Source Deep Learning Deployment Toolkit for Image-Guided Therapy},
  author={Mehrtash, Alireza and Pesteie, Mehran and Hetherington, Jorden and Behringer, Peter A and Kapur, Tina and Wells III, William M and Rohling, Robert and Fedorov, Andriy and Abolmaesumi, Purang},
  booktitle={Proceedings of SPIE--the International Society for Optical Engineering},
  volume={10135},
  year={2017},
  organization={NIH Public Access}
}