| |
Abstract:
Representations of the spatial structure of an individual
subject's cortical sheets are of anatomical interest, provide
important constraints for statistical analysis of functional
imaging data and aid their visualization. Due to noise and limited
resolution of anatomical magnetic resonance volumes, conventional
methods of cortex segmentation and reconstruction yield
representations that deviate from the cortical sheet's known simple
topology. The errors, called handles, have particularly deleterious
effects when inflations or flatmaps of the cortex are produced. So
far handles had to be removed by cumbersome manual editing, or
computationally very expensive methods of
reconstruction-by-morphing had to be used. Here we describe a
linear time complexity algorithm that automatically detects and
removes handles in voxel objects obtained by conventional
segmentation methods. The algorithm's core component is a region
growing process that starts deep inside the object, is prioritized
by the distance-to-surface of the voxels considered for inclusion
and is selftouching-sensitive, i.e. voxels whose inclusion would
add a handle are never included. The result is a binary voxel
object identical to the initial object except for "cuts" located in
the thinnest part of each handle. The accuracy of the resulting
representation of the cortical sheet is demonstrated visually, by
cross-validation between reconstructions from different scans of
the same subject's cortex and by comparison to solutions obtained
through manual intervention by an expert.
|
