Abstract— A new approach of registration in multimodal imaging has been developed. Modalities involved are Digital Subtracted Angiography (DSA, 2D) and Magnetic Resonance Angiography (MRA, 3D). Our approach is an hybrid one, mixing feature and intensity based approaches. This approach is based on the extraction of a anatomical referential common to both MRA and DSA. The results obtained prove the methods efficiency in a clinical context. This paper present the validation methodology to make it possible the replacement of the localization DSA examination by the diagnosis one, thus avoiding supplementary costs, lost time and medical hazards for the patient and for the medical staff. I. INTRODUCTION During the latest years, numerous methods of multimodal image matching have been developed. Associated with medical imaging, these developments make it possible to match images using intrinsic data, as anatomical data, instead of external referential, as stereotactic frames. Thus, the use of intrinsic matching considerably increases possibilities in medical image analysis. Unfortunately, these techniques mostly remain in the research field and are rarely used in clinical daily practice. In this paper, we present an original method for matching projective imaging (2D, radiography, angiography…) and tomographic imaging (3D, Magnetic Resonance Imaging, Computed Tomography). Furthermore, we propose a radiosurgical application as well as the different steps required to validate our approach and to replace the conventional technique using a stereotactic frame. We first briefly present registration of 2D and 3D data and the place of our solution among the existing techniques. Then, we deal with the needs of multimodal imaging in radiosurgery presenting the conventional data matching technique opposed to our 2D/3D intrinsic registration. A rigorous protocol of validation is proposed and the preliminary results of this study are presented, demonstrating the breakthrough of such methods in radiosurgery. II. MATERIAL AND METHODS Matching data can be shortly described as the computation of a mathematical transformation to place multimodality images (or volumes) in a single space to enable analysis or data fusion. Basically, those transformations are based on rotation, translation and scaling, even non-rigid deformations when needed (e.g. to compensate movements as respiration, bladder filling…). Data concerned can be either 2D-2D (images) or 3D-3D (volumes). In the case of 2D/3D data, a mathematical transformation has to describe the volume configuration in the space of the projective imaging ( Fig 1.) 3D Data 2D Reference 2D Floating Before registration 3D Data 2D Reference 2D Floating After registration Fig. 1. Basis of 2D/3D registration To obtain the transformation linking volume to projective image, algorithms are generally based on the same principle (see Fig. 2). To achieve registration, an optimization is performed: A) from its position in the projective space, the volume is projected to a plan, making a new 2D image (floating image) to be compared to the reference image B) a similarity measure estimates the difference between floating and reference images C) this difference is used to compute a new position of the volume. A 2D/3D matching based on a hybrid approach: improvement to the imaging flow for AVM radiosurgery M. Vermandel, N. Betrouni, D. Pasquier, J.Y. Gauvrit, C. Vasseur, J. Rousseau, Member, IEEE Proceedings of the 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference Shanghai, China, September 1-4, 2005 0-7803-8740-6/05/$20.00 ©2005 IEEE. 3071