Extracting and Representing the Cortical Sulci

IEEE Computer Graphics and Applications 49 Cortical sulci are convoluted regions between cortical folds deeply embedded in the surface of the brain. Because of their significant characteristics, their functional separation of distinct brain regions, and their natural topological partition of anatomy, sulci serve as a basis for structural analysis of the brain. Thus they receive significant attention in the neuroscience community. Sulcal features help define regions of the cortex and provide readily appreciated boundaries for teaching lobar and functional topography. However, the structural complexity of sulci poses a significant challenge for modeling, impeding rapid progress in quantitative morphological analysis. Figure 1 shows a typical sagittal 2D cross-section of magnetic resonance (MR) brain data. (A sagittal section lies parallel to the 2D plane separating the two brain hemispheres.) Recent research has made great progress in identifying and representing sulci, but so far automatic extraction and mathematical representation remain elusive. In this article we propose a robust voxel-coding methodology to automatically extract and parameterize the sulcal surfaces from volume data sets. We describe sulci as a collection of 2D contours generated from successive data slices. The contours are parameter curves of the sulci. The algorithm includes two phases: contour generation and sulcal surface generation. Phase 1 performs the segmentation, initial contour extraction, and final contour formation. The contours are 2D skeletons extracted from the segmented region. In contrast to those generated by traditional methods, our skeletons are one-pixel-thick curves starting with the exterior boundary, rather than a set of disconnected points. They thus provide desirable properties for further processing. The endpoints of each contour can be landmarks that represent it uniquely. In Phase 2, the integration between slices and parametric representation results from testing adjacency relationships of the contour endpoints within adjacent slices to resolve ambiguity in contour connections. The algorithm has the following features: