Impact of inter-subject image registration on group analysis of fMRI data

Abstract

A comparative study of three inter-subject brain image registration methods (SPM99, AFNI, and ART) is presented. It is shown that ART, which has a greater degree of freedom than SPM99 or AFNI, is able to more accurately remove the anatomical variability between high-resolution MR images of different subjects. The accuracy is assessed by the ability of the algorithm to reduce a measure of spatial dispersion among manually selected, homologous landmarks. We also investigated whether the superior ability of ART in removing inter-subject anatomical variance has any advantages for group analysis of functional magnetic resonance imaging (fMRI) data. In this study, data from a group of 21 subjects performing the visual oddball task were analyzed using three registration methods. The impact of inter-subject registration on the resulting activation maps was assessed using reproducibility and sensitivity measures derived from a nonparametric statistical analysis of the data. Using these measures, it is shown that a statistically significant increase in the reproducibility of activation maps and empirical sensitivity of activation detection can be achieved when ART is used for inter-subject registration. We conclude that there are significant advantages to be gained by using high dimensional, inter-subject registration methods for group analysis of fMRI data.