GEOMETRIC DISTORTION CORRECTION OF MAGNETIC RESONANCE IMAGES FROM DIFFERENT SITES AND SCANNERS

Abstract

One of the challenges of multi-site studies is to minimize image variability caused by geometric distortions. Past methods have focused on specially designed phantoms and gradient field maps for correction. This study aimed to create a correction for geometric distortion using a robust and validated nonlinear registration method to register T1-weighted structural brain scans of the same individuals from different scanners. Two strategies were proposed: the first was a direct averaging of the diffeomorphism across subjects, and the second approach included an additional step of transforming the diffeomorphisms onto an atlas space before averaging. Both were then evaluated with a leave-one-out approach. Using cortical thickness for validation, results showed that the correction applied made MRI data across scanners comparable; there was a decrease in mean difference between scans from the 2 scanners. The proposed correction would help to decrease scan time and costs associated with phantoms and field maps.