Data acquisition for Magnetic Resonance Imaging (MRI) is usually expensive and time-consuming. Multi-site study enables pooling more data with less cost. However, the reliability of multi-site study is not guaranteed since the data acquired from different sites always introduces site related variations. Further, these variation can not be fully resolved even using the same imaging protocols. In this thesis, we propose a seed-based image processing and statistical analyzing pipeline which mitigates the variations brought by sites to a statistically insignificant level. We collect data from a same group of subjects on two different scanners where each subject undergoes two imaging session on each site. Seed-based correlations of BOLD timeseries are used to access the connectivity between the human brain regions and seed region. The results imply that images collected from the four visits generate similar results of seed-based connectivity. The variance brought by site-related factors, machine, visit and interaction are proved to be insignificant by ANOVA test. Moreover, principle component analysis (PCA) are performed in a manner that data are reconstructed where subject identifiability is maximized. It is shown that reconstructed data introduces less variance from interaction of machine and visit.
History
Degree Type
Master of Science in Electrical and Computer Engineering