Evaluation of an edge-based registration method: application to magnetic resonance first-pass myocardial perfusion data Nadjia Kachenoura a,b, ⁎ , Philippe Cluzel a,b,c , Frédérique Frouin a,b , Dan Toledano c , Philippe Grenier a,b,c , Charles André Cuenod d,e,f , Daniel Balvay d,e a Inserm, U678, Laboratoire d'Imagerie Fonctionnelle, 75013, Paris, France b Université Pierre et Marie Curie, UMR_S 678, Laboratoire d'Imagerie Fonctionnelle, 75013, Paris, France c Assistance Publique-Hôpitaux de Paris, Pitié Salpêtrière Medical Center, Radiology Department, 75013, Paris, France d Inserm, U970, Paris Cardiovascular Research Center, 75015, Paris, France e Université Paris Descartes, 75015, Paris, France f Assistance Publique - Hôpitaux de Paris, European Hospital Georges Pompidou, Radiology Department, 75015, Paris, France Received 9 August 2010; revised 25 February 2011; accepted 26 February 2011 Abstract Purpose: Quantification of cardiac magnetic resonance (CMR) myocardial perfusion remains time consuming since it requires manual interventions to compensate for motion. Thus, the aim of this study was to evaluate a semiautomated registration method. Materials and Methods: A rigid edge-based registration algorithm was applied on 10 patients who had rest and stress CMR acquisitions on three slice levels (apical, midventricular and basal slices). Registration efficiency was assessed qualitatively by evaluating the quality of k-means maps in terms of symmetry and heart structures identification before and after registration and quantitatively by estimating noise amplitude within the myocardium. Finally, residual registration errors were manually estimated. Results: Before registration, k-means maps were satisfactory for 15 of 30 slices at rest and for only 5 of 30 slices during stress. After registration, the k-means maps quality was satisfactory for 29 of 30 slices at rest and for 30 of 30 slices during stress. Moreover, registration reduced noise amplitude from 49±26 to 29±11 at rest (Pb.01) and from 52±14 to 30±10 during stress (Pb.01). The residual horizontal and vertical shifts were 0.06±0.12 and 0.04±0.08 mm at rest and 0.32±0.69 and 0.28±0.53 mm at stress. Conclusion: The registration was successfully tested on rest and stress CMR perfusion data. It provides a valuable basis for quantitative evaluation of myocardial perfusion. © 2011 Elsevier Inc. All rights reserved. Keywords: Registration; Motion correction; Myocardial perfusion; Cardiac magnetic resonance 1. Introduction Cardiac magnetic resonance (CMR) myocardial perfusion imaging is an essential addition to the noninvasive cardiac imaging tools for the diagnosis and the follow-up of patients with coronary artery disease (CAD) [1–5]. With its high spatial and temporal resolutions and well-established techniques for cardiac gating and stress testing, it allows visualization of flow-limiting-related perfusion abnormali- ties and is thus increasingly used for the evaluation of CAD [2,6,7]. Several studies showed the potential usefulness of CMR myocardial perfusion imaging against conventional techniques used in clinical routine for the diagnosis and the evaluation of CAD such as positron emission tomography and single photon emission computed tomography myocar- dial perfusion imaging [8,9]. In clinical routine, first-pass magnetic resonance (MR) myocardial perfusion is commonly determined throughout visual analysis of dynamic contrast-enhanced sequences [1,5,10]. Several attempts for semiquantitative or quantita- tive analysis of CMR myocardial perfusion have been proposed and tested at rest and during pharmacological stress [2,5,6,11–15]. These techniques are mostly based on the analysis of the dynamic changes in MR imaging signal Available online at www.sciencedirect.com Magnetic Resonance Imaging 29 (2011) 853 – 860 ⁎ Corresponding author. Tel.: +33 1 53828437; fax: +33 1 53828448. E-mail address: nadjia.kachenoura@gmail.com (N. Kachenoura). 0730-725X/$ – see front matter © 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.mri.2011.02.020