Abstracts

Voxel-based morphometry (VBM) allows researchers to detect regional differences in gray or white matter volume between groups by spatially aligning subject brains to the same coordinate space such that voxel-wise comparisons can be made. Most VBM studies in the literature rely on techniques that automatically segment the brain into gray and white matter compartments, while simultaneously removing the skull and dura. These segmentation methods are not without error, and this error can contribute to an inaccurate spatial registration. The purpose of this paper is to show that images that have been more accurately segmented may offer superior results for VBM., VBM was performed on a previously investigated dataset consisting of 13 subjects with left temporal lobe epilepsy (TLE), 13 subjects with right TLE, and 61 control subjects (Neuroimage. 2004 Sep;23(1)167-74) to analyze regional differences in both gray and white matter. The SPM2 software was used to process and analyze all subject data. Subject input images were segmented using three different methods: 1) using the fully-automated template-based segmentation routine of SPM with default templates, 2) replacing the default templates with study-specific templates (SST-SPM), and 3) images segmented using the BRAINS2 software, which uses a multispectral segmentation routine to provide an intensity-based segmentation and is more accurate, but requires human intervention., The output of VBM differed depending on which segmentation method was used. Since both input images and spatial normalization routine were identical, the difference is dependent upon the segmentation method. Figure 1 shows decreases in white matter volume relative to controls in individuals with left TLE at a significance level of p [lt] 0.0005, uncorrected. While results are similar between each segmentation routine, the statistical output of the BRAINS2 segmented images in VBM exhibited the largest peak T-value of 8.13 in the temporal pole white matter compared to 6.82 and 6.03 in the SPM and SST-SPM segmented images, respectively. Note that volume changes in both SPM segmented groups near the corpus callosum and frontal white matter occurred near tissue compartment boundaries, and are most likely an artifact of tissue misclassification.[figure1], The increased effect apparent in the left TLE group in this study can be attributed to a more accurate spatial alignment within subjects. Therefore, it is advisable that when a more accurate segmentation method is available, it should be used to increase the validity of VBM results., (Supported by NIH NS 2RO1-37738 and NIH RO1 RR16591-02.)