Registration Of Myocardial Perfusion MR Image

Cardiac magnetic resonance (CMR) imaging is harmless to human body.Especially, the CMR perfusion imaging has been a hotpot in recent medicalimaging research for its value in cardiac functional measurements.However, in CMR image series, there is translation and deformationinduced by patient’s motion, respiration and heart moving. As a result, theyhurt the precision of the quantitative analysis. Therefore, image registrationwhich is a pre-processing for further image processing is introduced to tacklethis problem. Meanwhile, the image intensity will change continuously as thecontrast agent flowing in the heart, which make the intensity basedregistration difficult to be applied in perfusion image series. For the lowimage quality in the perfusion series, it’s hard to extract steadily features forfurther registration.In this paper, a registration algorithm which is a combination of imageenhancement, rigid and non-rigid registration is presented. Firstly, hybridimage enhancement is applied to the CMR image where the ventricular walland papillary muscles that are beneficial for image registration are enhancedin this step. Then normalized cross correlation (NCC) based rigid registrationis introduced to correct the large scale translation and rotation which isinduced by patient’s move and respiration. To implement the algorithmefficiently, a multi-resolution and hierarchical technique is introduced.Secondly, a markov random field (MRF) model is applied to correct thestill remaining elastic deformation. In the markov energy function, the imageblock is normalized to get rid of the effect of intensity change. Also, apiecewise continuous function is introduced to make the registration fieldsmooth. In order to remove the accumulated error which is common inregistration of image series, a pseudo ground truth (PGT) image is calculatedas template for each of the myocardial perfusion image. The accumulatederror can be reduced by warping each myocardial perfusion image to itscorresponding pseudo ground truth image.Detailed comparison and verification experiments demonstrate that ourmethod can effectively correct the displacement induced by patient’s moveand respiration, and deformation induced by heart-beat in the myocardialperfusion image. By using PGT image as template, accumulated error can bereduced.