Unsupervised identification of white matter tracts in a mouse brain using a directional correlation-based region growing (DCRG) algorithm

Abstract

The inner product of the major eigenvectors of adjacent pixels, known as the directional correlation (DC), has been used previously as a quantitative index to investigate directional similarity in white matter (WM) tracts. A high degree of directional similarity (i.e., high DC) among pixels within individual WM tracts was observed. Based on this observation, a region growing algorithm was employed to propagate an area from a seed point as a function of the DC threshold (DCt) to manually identify WM tracts in two-dimensional (2D) slices from diffusion tensor imaging (DTI). In the present study, an improved unsupervised DC based region growing (DCRG) method was implemented to reduce the operator-dependent variance and to improve the ease of use of the technique. By employing improved method, a multi-slice DTI data set of an in vivo mouse brain was used to identify the external capsule, visual pathway, and corpus callosum. The resultant WM tracts are computer rendered in three-dimensional (3D) images with anatomical images as structural references. In addition, three sets of ex vivo mouse brain data were used to examine the effects of different slice thickness and the signal-to-noise ratio (SNR) to the outcome of DCRG.

Introduction

Identification of white matter (WM) tracts by applying a directional correlation-based region growing (DCRG) method to diffusion tensor imaging (DTI) data has been reported previously (Hong and Song, 2001, Lin et al., 2003, Sun et al., 2001, Sun et al., 2003). The directional correlation (DC), defined as the inner product of the principle eigenvectors of adjacent pixels, is an index of directional similarity between those pixels. In a DTI data set, the adjacent regions within a WM tract possess high directional similarity in water diffusion. Thus, the WM tract of interest can be derived by selecting a seed point within the tract followed by repeated calculations of DC. The DCRG method allows the WM tract of interest to be identified and also systematically classifies other possible WM tracts within a two-dimensional (2D) image slice. Recently, the clinical application of this concept to human brain DTI data was reported (Klose et al., 2004). The authors successfully classified the central and lobar fibers employing a threshold of the magnitude of the angle between the principle eigenvectors of neighboring pixels. Both the approach of Klose et al. and the DCRG method rely heavily on the proper selection of the angular or DC threshold (DCt), which has previously been performed manually. To reduce the possible inconsistency and the laborious effort involved in manually selecting the threshold, the current study developed an unsupervised DCRG routine to select the optimized DCt.

Three-dimensional (3D) reconstruction of DTI data is commonly used to visualize WM tracts of interest (Horsfield and Jones, 2002, Kunimatsu et al., 2003, Mori et al., 2002, Pierpaoli et al., 2001, Simmons et al., 1999, Virta et al., 1999, Zhang et al., 2002). Not only can 3D presentation provide a more intuitive way of visualizing 3D objects such as WM tracts, it might also be useful in revealing morphological changes that are not obvious in 2D images. Thus, the extension of the single-slice DCRG method to multi-slice 2D or true-3D data sets was also focused on in the current study. An unsupervised procedure was established to select the optimized DCt values for a multi-slice data set. The identified WM tracts of interest were used to provide 3D volume-rendered visualizations of the tracts.