Factors affecting the voxel-based analysis of diffusion tensor imaging

• Published in: Chinese science bulletin Vol. 59; no. 31; pp. 4077 - 4085
• Main Authors: Wang, Jianli, Nie, Binbin, Zhu, Haitao, Liu, Hua, Wang, Jingjuan, Duan, Shaofeng, Shan, Baoci
• Format: Journal Article
• Language: English
• Published: Heidelberg Springer-Verlag 01.11.2014; Science China Press
• Subjects: Chemistry/Food Science; Earth Sciences; Engineering; Humanities and Social Sciences; image analysis; Life Sciences; microstructure; multidisciplinary; patients; Physics; Science; Science (multidisciplinary); seeds; t-test; 三线性插值; 体素; 可能影响; 成像; 扩散张量; 结构完整性; 统计分析; 预处理过程; Fractional anisotropy; Jaccard similarity; Voxel-based analysis; Statistical parametric mapping; Diffusion tensor imaging
• ISSN: 1001-6538; 1861-9541
• DOI: 10.1007/s11434-014-0551-8

Diffusion tensor imaging （DTI） provides a unique method to reveal the integrity of white matter microstructure noninvasively. Voxel-based analysis （VBA）, which is a highly reproducible and user-independent tech- nique, has been used to analyze DTI data in a number of studies. Fractional anisotropy （FA）, which is derived from DTI, is the most frequently used parameter. The parameter setting during the DTI data preprocessing might affect the FA analysis results. However, there is no reliable evidence on how the parameters affect the results of FA analysis. This study sought to quantitatively investigate the factors that might affect the voxel-based analysis of FA; these include the interpolation during spatial normalization, smoothing kernel and statistical threshold. Because it is difficult to obtain the true information of the lesion in the patients, we simulated lesions on the healthy FA maps. The DTI data were obtained from 20 healthy subjects. The FA maps were calculated using DTIStudio. We randomly divided these FA maps into two groups. One was used as a model patient group, and the other was used as a normal control group. Simulated lesions were added to the model patient group by decreasing the FA intensities in a specified region by 5 %- 50 %. The model patient group and the normal control groupwere compared by two-sample t test statistic analysis voxel- by-voxel to detect the simulated lesions. We evaluated these factors by comparing the difference between the detected lesion through VBA and the simulated lesion. The result showed that the space normalization of FA image should use the trilinear interpolation, and the smoothing kernel should be 2-3 times the voxel size of spatially normalized FA image. For lesions with small intensity change, FWE correction must be cautiously used. This study provided an important reference to the analysis of FA with VBA method.