Directional correlation characterization and classification of white matter tracts

Abstract

To study the architectural characteristics of white matter (WM) tracts, the directional correlation (DC), defined as the inner product of the major eigenvector of adjacent pixels, was used as a quantitative index to investigate directional similarity in WM tracts. A region-growing algorithm was employed to propagate an area from a seed point as a function of the DC threshold (DCt) to critically evaluate the directional properties of WM tracts. As the DCt was increased, more pixels were excluded from the propagated region as their DC fell below the DCt, and neighboring WM tracts could be distinguished as the area decreased. Taking the DC into account, a systematic classification routine for WM tracts was devised and tested on a mouse brain in vivo. The results show that individual WM tracts possess a high degree of directional similarity, and, by careful choice of the DCt value, the proposed classification algorithm can recognize all possible WM tracts in a given data set. Magn Reson © 2003 Wiley-Liss, Inc.