A complete pipeline for multi-atlas-based pituitary segmentation

Dependencies

• ANTs (https://github.com/ANTsX/ANTs)
• FreeSurfer (https://surfer.nmr.mgh.harvard.edu/)

Pipeline

1. Skull stripping using FreeSurfer's mri_watershed.
2. N4 bias correction using ANTs' N4BiasFieldCorrection.
3. Registration of user-provided atlases to the input image using ANTs' antsRegistration.
4. Registered masks are combined, a threshold is optionally applied to generate a binary mask.

Usage

Atlas-based pituitary segmentation using ANTs.

Usage: ./seg.sh <input> <output> [-a atlas_dir] [-t transform] [-s] [-n] [-m threads] [-h]

Options:
  <input>         Input image filename.
  <output>        Output image filename.
  -a atlas_dir    Path to the directory containing the atlas and mask images. Default: ./atlas.
  -t transform    Type of transform to use in registration. Default: SyNQuick. Currently supported: Affine, SyN, SyNQuick.
  -s              Supply this flag to SKIP skull stripping on the input image.
  -n              Supply this flag to SKIP N4 bias correction to the input image.
  -m threads      Number of threads to use. Default: 1. Increase this value to speed up the registration process.
  -h              Display this help message.

Example

For this pipeline, a couple of things are required:

1. A directory atlas_dir (which is specified using the -a flag) containing the atlas and mask images. Currently, exactly 10 atlases and corresponding masks are required. They should be organized as follows:

   - atlas_dir
       - 1_t1_strip_n4.nii.gz
       - 1_m_resampled.nii.gz
       - 2_t1_strip_n4.nii.gz
       - 2_m_resampled.nii.gz
       - ...
       - 10_t1_strip_n4.nii.gz
       - 10_m_resampled.nii.gz 
   

2. The atlas images should be skull-stripped and N4 bias corrected. The script to be used for preprocessing atlases is provided in preprocess.sh.

   Preprocess an image for pituitary segmentation.
   
   Usage: ./preprocess.sh <input> [-h]
   
   Options:
     <input>         Input image filename.
     -h              Display this help message.
   

   For example, calling

   ./preprocess.sh 1_t1.nii.gz
   

   will generate 1_t1_strip.nii.gz and 1_t1_strip_n4.nii.gz. One would then need to manually label the pituitary in 1_t1_strip_n4.nii.gz and save the label as 1_m_resampled.nii.gz.
3. Lastly, for an unprocessed image to be segmented, one would call the pipeline as follows:

   ./seg.sh input.nii.gz output.nii.gz -a atlas_dir
   

   The pipeline will then perform skull stripping, N4 bias correction, and atlas-based segmentation. The output will look like this:

   - input_strip.nii.gz
   - input_strip_n4.nii.gz
   - output_all.nii.gz
   - output_threshold_1.nii.gz
   - output_threshold_2.nii.gz
   - ...
   - output_threshold_10.nii.gz
   

   where output_all.nii.gz is the result of combining all registered masks, and output_threshold_X.nii.gz is the result of applying a threshold of X to the combined mask.
4. If the input image is already skull-stripped and N4 bias corrected, one can skip these steps by supplying the -s and -n flags, respectively.

   ./seg.sh input_strip_n4.nii.gz output.nii.gz -a atlas_dir -s -n
   

5. If unsatisfied with speed, one can increase the number of threads used by the pipeline.

   ./seg.sh input.nii.gz output.nii.gz -a atlas_dir -m 8
   

Different from antsJointLabelFusion.sh

• Additional (optional) pre-processing integrated into the pipeline.
• Option to use SyNQuick for faster registration.
• Does not contain Joint Label Fusion (JLF) step.