Difference between revisions of "Brain ventricle parcellation instructions"

From IACL
Jump to navigation Jump to search
 
Line 1: Line 1:
<meta name="title" content="Brain Ventricle Parcellation with Convolutional Neural Network"/>
+
<!-- <meta name="title" content="Brain Ventricle Parcellation with Convolutional Neural Network"/> -->
 
{{h2|Brain Ventricle Parcellation with Convolutional Neural Network}}
 
{{h2|Brain Ventricle Parcellation with Convolutional Neural Network}}
 
{{TOCright}}
 
{{TOCright}}

Latest revision as of 00:47, 3 July 2022

Brain Ventricle Parcellation with Convolutional Neural Network

This singularity image contains the code and trained model for brain ventricle parcellation using a convolutional neural network. The singularity image can be downloaded in the following link (~1.7GB): Singularity image for brain ventricle parcellation (v4).

If you use this work, please cite:

  • Shao, M., Han, S., Carass, A., Li, X., Blitz, A.M., Shin, J., Prince, J.L. and Ellingsen, L.M., 2019. Brain ventricle parcellation using a deep neural network: Application to patients with ventriculomegaly. NeuroImage: Clinical, p.101871.

If you have any questions, please email Muhan Shao at muhan@jhu.edu.

Processing steps

The singularity takes T1-w RAW brain MRI (nifti file) as input and performs the following steps:

  • N4 bias field correction from ANTs. The bias field is estimated using a weight image calculated from a brain mask generated by ROBEX.
  • Rigid registration to ICBM2009c nonlinear symmetric template using the ANTs package. The template images were resampled to have resolution of 0.8x0.8x0.8mm.
  • Skull-stripping using ROBEX.
  • Brain ventricle parcellation on the skull-stripped MNI-registered T1-w MRI using the method described in "Shao, M., et al., 2019. Brain ventricle parcellation using a deep neural network: Application to patients with ventriculomegaly. NeuroImage: Clinical, p.101871".
  • The parcellation is transformed back to the original space using ANTs with the "MultiLabel" interpolation.

Output file structures

The processing will create subfolders n4/, mni/, and parc/ under the output folder. The final parcellation result is directly under the output folder.

  • *n4.nii.gz is the bias field corrected image
  • *n4_mni.nii.gz is the image in the MNI space
  • *n4_mni_strip.nii.gz is the skull-stripped image
  • *n4_mni_strip_seg.nii.gz is the parcellation from the convolutional neural networks
  • *n4_mni_strip_seg_inverse.nii.gz is the final parcellation result in the original image space.

Installation

Usage

  • The Singularity image can only run on CPU although TensorFlow and Keras are used, because the Singularity image only contains the CPU version of TensorFlow.
  • Assume that the Singularity image is /path/to/simg/ventricle-parcellation.simg, the image to parcellate is ~/image.nii.gz, and the output folder is ~/output
singularity run /path/to/simg/ventricle-parcellation.simg -i ~/image.nii.gz -o ~/output
  • If the image.nii.gz is under /path/to/image which is not under your home directory
singularity run -B /path/to/image:/mnt /path/to/simg/ventricle-parcellation.simg -i /mnt/image.nii.gz -o /mnt/output
  • Print help
singularity run ~/ventricle-parcellation.simg -h

Brain ventricle labels

   "51":  "Right lateral ventricle",
   "52":  "Left lateral ventricle",
   "4":  "third ventricle",
   "11":  "fourth ventricle"