BrainVoyager v23.0
Overview of Cortex Segmentation Steps
Achieving accurate segmentation of the cortical sheet is a very difficult task, which involves a substantial amount of work if done manually. In this section a set of tools is described with the goal to create correct segmentations of the cortical sheet of the left and right hemisphere in a largely automatic way. The better the grey / white matter contrast of the used 3D MRI data sets, the less manual work is necessary. It is recommended to use optimized structural T1-weighted MRI sequences in order to reduce manual corrections to a minimum. The automatic segmentation procedure runs through a sequence of steps including a filter enhancing tissue contrast (sigma filter), masking to remove the cerebellum and non-brain tissue, filling of ventricles and subcortical structures, intensity histogram analysis to calculate proper thresholds to separate white and grey matter, morphological operations to smooth the resulting segmentation and a "bridge removal" tool (contributed by Niko Kriegeskorte) to remove topological errors.
The automatic segmentation pipeline assumes that the input data is a (iso-voxeled) 1 mm data set in ACPC or Talairach space. If you have data that was truly acquired (without interpolation) at sub-millimeter voxel resolution, you may consider using the advanced segmentation tools. Since BrainVoyager QX 2.8 it is also possible to run the automatic segmentation pipeline using the advanced segmentation routines for 1 mm data sets by turning on the High-resolution (upsampled 0.5 mm iso voxel) option in the Automatic Cortex Segmentation and Reconstruction dialog.
Prerequisites
The following prerequisites should be considered before running the automatic segmentation tools:
- The original anatomical 3D data set containing the brain must have an appropriate spatial resolution, i.e. close to 1mm x 1mm x 1mm. If the voxels do not have exactly a resolution of 1mm, the iso-voxel tool should be used to interpolate the data set. If header information is available, interpolation to 1mm iso-voxels is suggested automatically by the program.
- The contrast between grey matter and white matter tissue must be high enough. We recommend the "ADNI" 3D MPRAGE sequence for high-quality anatomical scanning since this sequence is optimized for white/gey matter contrast and because it is available for scanners of most manufacturers; if not available, a conventional MPRAGE or similar sequence with high white / grey matter contrast should be used. Averaging multiple structural scans of the same subject further improves the quality of the data. If multiple high-quality scans of the same subject have been acquired in different scanning sessions, the 3D-3D (VMR-VMR) alignment tool in the Coregistration tab of the 3D Volume Tools dialog can be used; after aligning the two (or more) anatomical data sets they can bre averaged using the average data sets tool in the Combine 3D Data Sets dialog.
- The intensities across space of the different tissue types should be as homogenous as possible. It is strongly advised to run the automatic intensity inhomogeneity correction tool before performing cortex segmentation; this tool should be run before ACPC/Talairach transformation to ensure that the original 16-bit (.V16) data is used (alternatively, the 16-bit data can also be brought in ACPC/Talairach space). If the displayed (final) histogram of the intensity inhomogeneity correction tool does not show two clearly separated peaks for white and grey matter, cortex segmentation will not be possible.
- While not strictly necessary, the brain should be segregated from the surrounding head tissue. Note that this "brain peeling" is included in the automatic intensity inhomogeneity correction step mentioned above but if this tool is not used, the brain extraction step should be performed explicitly.
- The anatomical data set should be transformed into ACPC or Talairach space because some tools of the automatic segmentation procedure exploit anatomical knowledge for initial brain segmentation (Talairach masking), filling of ventricles, removing subcortial structures and disconnection of cortical hemispheres. Use the spatial transformation tools to transform your original inhomogeneity corrected anatomical data set into ACPC or Talairach space.
The automatic segmentation routines attempt to provide high-quality results. Still it is recommended that you check the segmentation result using provided visualization tools and to improve it if necessary using provided correction ("drawing") tools.
Copyright © 2023 Rainer Goebel. All rights reserved.