volume_patches
legacy non-local patch-based segmentation using MINC1 I/O
volume_patches <input> [output_labels] [options]
DESCRIPTION
volume_patches performs non-local patch-based segmentation using MINC1 I/O. It compares local image patches between the input volume and a set of training volumes (atlases with known labels) to determine the most likely label for each voxel.
The tool uses a non-local means approach where each voxel’s label is determined by a weighted vote from similar patches found in the training library. Patch similarity is computed within a search neighbourhood, and the weights are derived from the patch distance using an exponential kernel controlled by the beta parameter.
This is the legacy MINC1 implementation. For the more capable ITK-based version that supports additional label fusion methods and file formats, see itk_patch_segmentation.
OPTIONS
--train <file>- Specify a training library file listing atlas volumes and their labels.
--train2 <file>- Specify a second training library for two-level segmentation.
--verbose- Print detailed progress information.
--quiet- Suppress all output messages.
--clobber- Overwrite existing output files.
--mask <file>- Specify a mask to restrict segmentation to a region of interest.
--patch <n>- Set the patch radius in voxels (default: 1).
--search <n>- Set the search radius in voxels for finding matching patches (default: 1).
--dist- Output the distance map instead of labels.
--cls <n>- Number of classes for segmentation.
--beta <value>- Regularization parameter controlling the exponential weighting kernel (default: 1.0).
--scaling- Enable intensity scaling between input and training volumes.
EXAMPLES
Segment using a training library:
volume_patches input.mnc output_labels.mnc --train library.txt --mask brain_mask.mnc
Segment with larger patch and search radii:
volume_patches input.mnc labels.mnc --train library.txt --patch 2 --search 2
Segment with custom beta and verbose output:
volume_patches input.mnc labels.mnc --train library.txt --beta 0.5 --verbose
AUTHOR
Vladimir S. Fonov - McConnell Brain Imaging Centre, Montreal Neurological Institute, McGill University.
COPYRIGHTS
Copyright © 2012 by Vladimir S. Fonov