Comparison of tract-based spatial statistics and ROI-based approach in analyzing the white matter integrity in the elderly

Abstract

The purpose of this study was to compare tract-based-spatial-statistics (TBSS) and ROI-based approach in examining the white matter integrity between AD, MCI, and normal control (NC) subjects who have no obvious WM lesions. Sixty subjects without obvious white matter lesions were selected from a total of 243 subjects by two neuroradiologists. DTI images were acquired with 64 encoding directions using a Siemens 3T MRI scanner. The demographic characteristics were matched among the 3 groups (AD: n = 19; M/F = 7/12, 75.3 ± 5 years; MCI: n = 18; M/F = 9/9, 73.2 ± 3.7 years; NC: n = 23; M/F = 7/16, 73.6 ± 5.9). TBSS analysis was performed with the FSL software package from Oxford's Analysis Group. With the ROI-based approach, fractional anisotropy (FA) was calculated as the standard deviation of eigenvalues from the mean eigenvalue normalized by square norm of eignvalues. With TBSS analysis, no significant differences were found in white matter tracts when AD was compared with MCI and MCI was compared with NC. Relative to NC, AD subjects exhibited significant lower FA values in the neural tracts of corpus callosum body, particularly in its radiations of the forceps major and forceps minor, which are categorically anterior and posterior cingulate fibers, respectively (Fig.1). There were also significant FA differences in the inferior and superior longitudinal fasciculus (p < 0.02). The ROI-based approach showed that the MCI group had significantly lower FA values in parietal WM (p < 0.001) compared to NC; and the AD group had significantly lower FA values in parietal, temporal, frontal WM, parahippocampal and posterior cingulate fibers (p < 0.001 for all). Compared to MCI, the AD group had significantly lower FA values in parietal (p = 0.005), temporal, frontal WM, and parahippocampal and posterior cingulate fibers (p < 0.001 for all). The results may suggest that although TBSS uses WM tract skeleton for spatial normalization, the difference in small areas may be lost. The ROI based analysis uses native image, and may be more sensitive to detect them; however, it is time consuming and not suitable for analyzing differences in the whole brain. More research is needed to improve the group comparison analysis for DTI.