ITK patch morphology tools: denoising, segmentation, grading

implements methods from:

• denoising:

  • Pierrick Coupé, Pierre Yger, Christian Barillot "Fast Non Local Means Denoising for 3D MR Images" http://dx.doi.org/10.1007/11866763_5

• adaptative denoising:

  • José V. Manjón PhD,Pierrick Coupé PhD, Luis Martí-Bonmatí PhD, D. Louis Collins PhD and Montserrat Robles PhD "Adaptive non-local means denoising of MR images with spatially varying noise levels" http://dx.doi.org/10.1002/jmri.22003

• segmentation:

  • Pierrick Coupé, , José V. Manjón, Vladimir Fonov, Jens Pruessner, Montserrat Robles, D. Louis Collins "Patch-based segmentation using expert priors: Application to hippocampus and ventricle segmentation" http://dx.doi.org/10.1016/j.neuroimage.2010.09.018

  • "MICCAI 2012 Workshop on Multi-Atlas Labeling" ISBN-10: 1479126187 entry BIC-IPL and BIC-IPL-HR, https://masi.vuse.vanderbilt.edu/workshop2012/images/c/c8/MICCAI_2012_Workshop_v2.pdf

  • Katrin Weier, Vladimir Fonov, Karyne Lavoie, Julien Doyon and D. Louis Collins "Rapid automatic segmentation of the human cerebellum and its lobules (RASCAL) Implementation and application of the patch-based label-fusion technique with a template library to segment the human cerebellum" http://dx.doi.org/10.1002/hbm.22529

• grading:

  • Pierrick Coupé, Simon F Eskildsen, José V Manjón, Vladimir S Fonov, D Louis Collins, Alzheimer's disease Neuroimaging Initiative "Simultaneous segmentation and grading of anatomical structures for patient's classification: application to Alzheimer's disease" http://dx.doi.org/10.1016/j.neuroimage.2011.10.080

Building:

• Building require ITK version 4.XX and cmake version 2.6 or later
• legacy tool ( volume_patches ) requires LIBMINC ( https://github.com/BIC-MNI/libminc )

Usage (concise):

These programs were originally designed to be used with MINC files, but should work with any file format supported by ITK version 4 Most tools have an optional paramter called search radius - which specifies how non-local the search should be (in voxels) and patch radius - the radius of the local patch used to extract features.

Denoising

Denoising requires specifying noise level ($sigma), and optionall search radius $search_radius and patch radius $patch_radius

itk_minc_nonlocal_filter input.mnc output.mnc --noise $sigma --search $search_radius --patch $patch_radius

Adaptative denoising

Adaptative denoising have optional parameters: search radius $search_radius and patch radius $patch_radius

itk_minc_nonlocal_filter input.mnc output.mnc --search $search_radius --patch $patch_radius --anlm

Segmentation

All segmentation tools and scripts require library of labelled samples.

Segmetnation tool require library of presegmented examples $train, number of classes including background $classes and optionally search radius and patch radius Training examples are referenced in a comma separated file in the format: <image.mnc>,<labels.mnc>

itk_patch_morphology input.mnc output_labels.mnc --discrete $classes --search $search_radius --patch $patch_radius --train $train

Several high level scripts are included in scripts directory:

• ventricles_segmentation_pipeline.pl - segmentation script for latera ventricle segmentation, uses volume_patches program from legacy directory
• hcag_segmentation_pipeline.pl - Hippocampus and Amygdala segmentation script
• miccai2012_segmentation_minipipe.pl - whole head segmentation pipeline, used in MICCAI 2012 Grand Challenge and Workshop on Multi-Atlas Labeling
• patch_segmentation_pipeline.pl - generic segmetnation script used in RASCAL paper

Grading

All grading tools and scripts require library of labelled samples.

Similar to segmetnation tool, grading require library of presegmented examples $train and optionally search radius and patch radius Training examples are referenced in a comma separated file in the format: <image.mnc>,<labels.mnc>,<grading> Two training libraries can be provided , which are both loaded (and optionally each is used independently for pre-selection)

itk_patch_morphology input.mnc --grading output_grading.mnc  --search $search_radius --patch $patch_radius --train $train --train2 $train2

High level script for simultaneous grading and segmentation: scripts/snipe_grading_pipeline.pl