IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 22, NO. 11, NOVEMBER 2003 1417 Intensity-Based 2-D–3-D Registration of Cerebral Angiograms John H. Hipwell*, Graeme P. Penney, Robert A. McLaughlin, Kawal Rhode, Paul Summers, Tim C. Cox, James V. Byrne, J. Alison Noble, and David J. Hawkes Abstract—We propose a new method for aligning three-di- mensional (3-D) magnetic resonance angiography (MRA) with 2-D X-ray digital subtraction angiograms (DSA). Our method is developed from our algorithm to register computed tomography volumes to X-ray images based on intensity matching of digitally reconstructed radiographs (DRRs). To make the DSA and DRR more similar, we transform the MRA images to images of the vas- culature and set to zero the contralateral side of the MRA to that imaged with DSA. We initialize the search for a match on a user defined circular region of interest. We have tested six similarity measures using both unsegmented MRA and three segmentation variants of the MRA. Registrations were carried out on images of a physical neuro-vascular phantom and images obtained during four neuro-vascular interventions. The most accurate and robust registrations were obtained using the pattern intensity, gradient difference, and gradient correlation similarity measures, when used in conjunction with the most sophisticated MRA segmenta- tions. Using these measures, 95% of the phantom start positions and 82% of the clinical start positions were successfully registered. The lowest root mean square reprojection errors were 1.3 mm (standard deviation 0.6) for the phantom and 1.5 mm (standard deviation 0.9) for the clinical data sets. Finally, we present a novel method for the comparison of similarity measure performance using a technique borrowed from receiver operator characteristic analysis. Index Terms—2-D–3-D registration, digital subtraction angiog- raphy, magnetic resonance angiography, neuro-interventions, sim- ilarity measures. I. INTRODUCTION M INIMALLY invasive neuro-interventions such as the treatment of arteriovenous malformations (AVMs) or intracranial aneurysms via endovascular coiling or glue embolization [1], [2], routinely make use of fluoroscopy and digital subtraction angiography (DSA) to guide the placement of instruments, such as catheters and coils. However the two-di- Manuscript received December 18, 2001; revised June 29, 2003. This work was supported in part by the EPSRC under Grant GR/M55015 and Grant GR/M55008. Asterisk indicates corresponding author. *J. H. Hipwell is with the Division of Imaging Sciences, UMDS, Guy’s & St Thomas’ Hospitals, London SE1 9RT, U.K. (e-mail: john.hipwell@kcl.ac.uk). G. P. Penney, K. Rhode, and D. J. Hawkes are with the Division of Imaging Sciences, UMDS, Guy’s & St Thomas’ Hospitals, London SE1 9RT, U.K. P. Summers is with the Institute of Neuroradiology, University Hospital Zurich, 8091 Zurich, Switzerland. R. A. McLaughlin and J. A. Noble are with the Medical Vision Laboratory, Department of Engineering Science, University of Oxford, Oxford OX2 7BZ, U.K. T. C. Cox is with the National Hospital for Neurology and Neurosurgery, Radiology Department, London WC1N 3BG, U.K. J. V. Byrne is with Department of Radiology, University of Oxford, The Rad- cliffe Infirmary, Oxford, Oxford OX2 6HE, U.K. Digital Object Identifier 10.1109/TMI.2003.819283 mensional (2-D) nature of these images makes navigation of the tortuous cerebral vessels a demanding task and complicates, in the case of endovascular coil embolization for instance, the delivery of coils through the neck of the aneurysm. Reg- istration of the 2-D X-ray projection images to preoperative magnetic resonance angiography (MRA) can greatly enhance visualization and introduces potentially useful complementary information such as three-dimensional (3-D) blood flow. There have been a number of papers describing 2-D–3-D registrations of MRA and X-ray angiographic images [3]–[6]. These studies have favored a feature-based approach in which, for instance, 2-D and 3-D vascular skeletons are extracted and compared using a suitable distance metric. While these approaches exhibit rapid execution times and high robustness in the face of large misregistrations, their reliance on accurate segmentations can limit the accuracy of the registration. In comparison, intensity-based algorithms are not dependent upon accurate segmentations and have been shown for other applications to give substantially more accurate and reliable results than some feature-based approaches [7], [8]. Intensity-based algorithms have not to date been applied suc- cessfully to the registration of 3-D MR angiograms with 2-D X-ray angiograms. Early attempts to do this failed due to the relative sparseness of the linear structures depicting the vascula- ture in both the 2-D and 3-D angiographic images. In this paper, we take as our starting point the algorithm described by Penney et al. [9] and develop a method that is both robust and accurate. In summary, we preprocess the MR angiograms; first, to re- move any contralateral vasculature to that imaged with DSA. We then produce an image of the vasculature in order to gen- erate a projection image in which the intensity of each pixel is approximately proportional to the thickness of vessel lumen tra- versed. This projection, analogous to the digitally reconstructed radiograph (DRR) in [9], and subsequently referred to as such, is matched directly to the X-ray DSA image. To improve robust- ness of the search for alignment we define two concentric re- gions of interest (ROIs) around the vasculature of interest. This identifies one point (the center of the sphere and its projection into 2-D) which acts as a starting point of the registration and re- stricts the registration algorithm to structures with shapes more amenable to a multiresolution search. In order to determine the most appropriate similarity measure for this new application, we compare the performance of six commonly used similarity measures when applied to the registration of both phantom and routinely acquired clinical images. In Section II, we describe the algorithm in more detail to- gether with developments required to tailor it to operate with 0278-0062/03$17.00 © 2003 IEEE