Cortical reconstruction using implicit surface evolution: A landmark validation study

Abstract

A validation study was conducted to assess the accuracy of an algorithm developed for automatic reconstruction of the cerebral cortex from T1-weighted magnetic resonance (MR) brain images. Manually selected landmarks on different sulcal regions throughout the cortex were used to analyze the accuracy of three reconstructed nested surfaces - the inner, central, and pial surfaces. We conclude that the algorithm can find these surfaces with subvoxel accuracy, typically with an accuracy of one third of a voxel, although this varies by brain region and cortical geometry. Parameters were adjusted on the basis of this analysis in order to improve the algorithm's overall performance. 1 © Springer-Verlag Berlin Heidelberg 2004.