anisotropic_nets_brats_challenge

Automatic brain tumor segmentation on BRATS with anisotropic nets

This page describes how to acquire and use the whole tumor segmentation network as a part of the pipeline described in:

Wang et al., Automatic Brain Tumor Segmentation using Cascaded Anisotropic Convolutional Neural Networks, MICCAI BRATS 2017

https://arxiv.org/abs/1709.00382

• This implementation ranked the first (in terms of averaged Dice score 0.90499) according to the online validation leaderboard of BRATS challenge 2017.*

• For a full implementation of the method described in this paper with three stages of the cascaded CNNs, please see: https://github.com/taigw/brats17

Downloading model zoo files

The network weights and examples data can be downloaded with the command

net_download anisotropic_nets_brats_challenge_model_zoo

(Replace net_download with python net_download.py if you cloned the NiftyNet repository.)

After running this command successfully, the downloaded files include:

• brats_seg_app.py -- an application built with NiftyNet, defines the main workflow of network training and inference.
• wt_net.py -- the network definitions.
• .ini files -- configuration files define system parameters for running segmentation networks, the three files correspond to the networks in three orientations [axial, coronal, sagittal] configurations.
• label_mapping_whole_tumor.txt -- mapping file used by NiftyNet, to convert the multi-class segmentations into a binary problem.

Generating segmentations for example data

Generate segmentations for the included example image with the command,

For the network operates in axial view:

net_run inference -a anisotropic_nets_brats_challenge.brats_seg_app.BRATSApp \
                  -c ~/niftynet/extensions/anisotropic_nets_brats_challenge/whole_tumor_axial.ini

For the network operates in coronal view:

net_run inference -a anisotropic_nets_brats_challenge.brats_seg_app.BRATSApp \
                  -c ~/niftynet/extensions/anisotropic_nets_brats_challenge/whole_tumor_coronal.ini

For the network operates in sagittal view:

net_run inference -a anisotropic_nets_brats_challenge.brats_seg_app.BRATSApp \
                  -c ~/niftynet/extensions/anisotropic_nets_brats_challenge/whole_tumor_sagittal.ini

(Replace net_run with python net_run.py if you cloned the NiftyNet repository.)

Generating averaged volume from the outcomes of the previous step

A script has been created to compute the averaged volumes and the Dice coefficients from the probabilistic outputs of the segmentation step.

python ~/niftynet/extensions/anisotropic_nets_brats_challenge/average_volume.py

A script rename_crop_BRATS.py is also available to to rename BRATS17 images into TYPEindex_modality.nii.gz format and crop with a bounding box to remove image background (voxels with intensity value zero).

---

Example data used in this model zoo entry are taken from Multimodal Brain Tumor Segmentation Challenge 2017.

Data references:

Menze, Bjoern H., et al. "The multimodal brain tumor image segmentation benchmark (BRATS)." IEEE transactions on medical imaging 34.10 (2015): 1993-2024. DOI: 10.1109/TMI.2014.2377694

Bakas, S., et al. "Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features." Nature Scientific Data 4:170117 (2017). DOI: 10.1038/sdata.2017.117

---

This model zoo entry is licensed under a Creative Commons Attribution 4.0 International (CC BY) License.