A Comprehensive Gaussian Process Framework for Correcting Distortions and Movements in Diffusion Images

Jesper L. R. Andersson¹, Junquian Xu², Essa Yacoub², Edward Auerbach², Steen Moeller², Kamil Ugurbil²

Registration based approaches to correcting for eddy current distortions and movements are complicated by the images containing different information. Possible solutions to this is i) To collect pairs of dwis with opposite polarity diffusion gradients, ii) To collect pairs of dwis with opposite polarity phase-encoding or iii) To register observed dwis to model based predictions. We present a method that include any and all of those sources of information depending on how the data was collected. It utilises Gaussian Processes to predict dwis and to predict eddy currents for one acquisition given other acquisitions.