Tong Zhu1, Rui Hu2, Xing Qiu2, Wei Tian1, Sven Ekholm1, Jianhui Zhong1
1Imaging Sciences, University of Rochester, Rochester, NY, United States; 2Biostatistics and Computational Biology, University of Rochester, Rochester, NY, United States
Diffusion tensor imaging (DTI) technique has been widely applied to study white matter (WM) abnormality longitudinally. For many neurodegenerative diseases, such as multiple sclerosis (MS), the WM abnormality has strong subject- or time-dependent heterogeneous patterns towards which most group-based analyses for DTI, including VBM and TBSS, are less sensitive. In this study, we propose a novel statistical analysis approach for DTI based on a nonparametric spatial regression fitting of DTI data among neighboring voxels. Statistical inference can be made for both group comparison among individuals and longitudinal comparison within the same individual. Effectiveness of this approach on group comparison was compared with the VBM approach while the proof of concept for longitudinal analysis of a single subject was evaluated through a MS patient with progressive lesions. Results show that the new method is comparable to the VBM approach and performs better than VBM when the number of scans per group is small than 5. Moreover, it detected longitudinally decreased FA from only one DTI scan per time point for a MS patient with progressive lesions. Our method provides a potentially better way to assess individual abnormality to determine more directly how the brain has changed as a result of disease/injury.