Representing the shape of the human cerebral cortex arises as a basic subproblem in several areas of brain science, such as when describing the anatomy of the cortex and when relating functional measurements to cortical regions.
Most current methods for building such representions of the cortical surface are either based on contours from two-dimensional cross sections or landmarks that have been obtained manually.
In this article, we outline a methodology for semi-automatic contruction of a solely surface based representation of the human cerebral cortex in vivo for subsequent generation of (unfolded) two-dimensional brain maps.
The method is based on input data in the form of three-dimensional NMR images, and comprises the following main steps:
- suppression of disturbing fine-scale structures by linear and non-linear scale-space techniques,
- generation of a triangulated surface representation based on either iso-surfaces or three-dimensional edge detection,
- division of the surface model into smaller segments based on differential invariants computed from the image data.
When constructing an unfolded (flattened) surface representation, the instrinsic curvature of the cortex means that such a unfolding cannot be done without introducing distortions. To reduce this problem, we propose to cut the surface into smaller parts, where a ridge detector acts as guideline, and then unfold each patch individually, so as to obtain low distortions.
Having a solely surface based representation of the cortex and expressing the image operations using multi-scale differential invariants in terms of scale-space derivatives as done in this work is a natural choice both in terms of conceptual and algorithmic simplicity. Moreover, explicitly handling the multi-scale nature of the data is necessary to obtain robust results.
1996. S126-S126 p.
QC 20111102 NR 20140804. NR 20160314