Eloy Roura1,
Torben Schneider2, Marc Modat3, Pankaj Daga3,
Nils Muhlert2, Declan T. Chard2, Sebastien Ourselin3,
Xavier Llad1, Claudia Angela M. Wheeler-Kingshott2
1Dept.
of Computer Architecture and Technology,
VICOROB, University of Girona, Girona, Spain; 2NMR Research
Unit, Queen Square MS Centre, Dept. of Neuroinflammation UCL Institute of
Neurology, London, United Kingdom; 3Centre for Medical Image
Computing, Dept. of Medical Physics and Bioengineering, UCL, London, United
Kingdom
In this work, we tested the performance of a registration framework to simultaneously register T1-weighted (T1w) and diffusion weighted (DW) images to a target space, comparing the results of this multi-channel (MC) approach against two different strategies: i) single channel (SC) (the T1w and FA separately); ii) T1w-based (using only the T1w SC registration results applied to both T1w and FA volumes). We generated 100 simulated MS patients by deforming 10 healthy subjects into 10 MS patients. Experimental results were assessed with both qualitatively using difference image and checkerboards and quantitatively computing the mean intensity of the difference image. MC approach yielded better results rather than SC or using the T1w-based transformations.
Keywords