Pseudo ground truth based nonrigid registration of myocardial perfusion MRI

This paper presents a novel nonrigid registration method for myocardial perfusion magnetic resonance (MR) images. To overcome the rapid intensity change due to contrast enhancement, we propose to register the observed sequence to a pseudo ground truth, which is a motion/noise free sequence that is estimated from the observed one, and having almost identical intensity variations as the original sequence. The pseudo ground truth and the elastic deformation fields for the observed sequence are obtained by minimizing an energy functional integrating both the registration error and the spatiotemporal constraints on the pseudo ground truth in an expectation-maximization framework. We have tested the proposed nonrigid registration method on 20 cardiac perfusion MR scans. The proposed method successfully compensated the elastic deformation of the heart in most scans according to visual validation. For quantitative validation, we propagated manually drawn myocardial contours in one frame to other frames according to the deformation fields obtained by applying different registration methods. The root mean square distance between the propagated contour and the gold standard is 2.11 mm if only global translation is compensated, and 1.87 mm after nonrigid registration, as compared with 2.80 mm for serial demons registration and 2.77 mm for a free-form deformation approach using normalized mutual information as the similarity measure, both of which adversely increased the error due to misregistration.

(1) Pseudo ground truth estimation by using conjugate gradient method. (2) Corresponding image registration by using edge-emphasized demons.Figure optionsDownload high-quality image (86 K)Download as PowerPoint slideResearch highlights
► Overcomes the rapid intensity change due to contrast enhancement.
► Incorporates segmentation results to improve pseudo ground truth estimation.
► Requires no pharmacokinetic model or arterial supply function.
► Outperforms serial demons and MI-based FFD approaches.