2D-3D intensity based registration of DSA and MRA – A comparison of similarity measures

© Springer-Verlag Berlin Heidelberg 2002. We have compared the performance of six similarity measures for registration of three-dimensional (3D) magnetic resonance angiography (MRA) to two-dimensional (2D) x-ray angiography images of the cerebral vasculature. The accuracy and robustness of each measure was investigated using a ground truth registration of a neuro-vascular phantom which was obtained using fiducial markers, and using “gold-standard” registrations of four clinical data sets calculated using manual alignment by a neuro-radiologist. Of the six similarity measures, pattern intensity, gradient difference and gradient correlation performed consistently accurately and robustly for all data sets. Using these similarity measures, and for starting positions within 8° rotation, 3mm in-plane translation and 50mm out-of-plane translation from the gold-standard/ground-truth positions, we obtained a success rate of greater than 80% for the clinical data sets, whilst none of the phantom registrations failed. The root-mean-square (rms) target reprojection error was less than 1.3mm for the clinical data sets. The rms target reprojection error for the phantom images was less than 1mm when using the most accurate similarity measures.