Reducing inter-scanner variability in multi-site fMRI data: Exploring choice of reference activation map and use of correction functions

Abstract

There has been an exponential growth in brain mapping studies in the past decade using functional MRI (fMRI). Apart from simple fMRI studies from a single site (scanner), multi-site studies are gaining great attention, as it has the potential to provide more data for brain mapping studies, thereby increasing the statistical power of the brain mapping studies. Major limitations with the multi-center studies are the diversity in acquisition and analysis methods, which affect the imaging results. This study aims at finding a suitable standard reference map for the sensori-motor fMRI task for the multisite fMRI data. After the construction of a suitable imaginary standard reference map, the study further tries to reduce the inter-scanner differences in multi-center fMRI data by using correction functions. It is observed that the median images (an imaginary reference image) work as a good reference map, than the reference images selected from the data set. A simple linear regression based correction estimate to the new reference map is seen to improve the classification accuracy of the functional activation maps, thus reducing the inter-scanner variability in multi-site fMRI data.