Left Ventricular Segmentation in MR Using Hierarchical Multi-class Multi-feature Fuzzy Connectedness Amol Pednekar 1 , Uday Kurkure 1 , Raja Muthupillai 2 , Scott Flamm 3 , and Ioannis A. Kakadiaris 1⋆ 1 Visual Computing Lab, Dept. of Computer Science, University of Houston, Houston, TX, USA 2 Philips Medical Systems North America, Bothell, WA, USA 3 Dept. of Radiology, St. Luke’s Episcopal Hospital, Houston, TX, USA Abstract. In this paper, we present a new method for data-driven au- tomatic extraction of endocardial and epicardial contours of the left ven- tricle in cine bFFE MR images. Our method employs a hierarchical, multi-class, multi-feature fuzzy connectedness framework for image seg- mentation. This framework combines image intensity and texture infor- mation with anatomical shape, while preserving the topological relation- ship within and between the interrelated anatomical structures. We have applied this method on cine bFFE MR data from eight asymptomatic and twelve symptomatic volunteers with very encouraging qualitative and quantitative results. 1 Introduction Magnetic resonance imaging (MRI) is the preferred cardiac imaging technique as it acquires images in oblique planes; obviating geometric assumptions regard- ing the shapes of the ventricles providing high blood-to-myocardium contrast. However, the cine cardiac MR (CMR) images are not incisive; they are fuzzy due to patient motion, background variation, and partial voluming. Left ventri- cle (LV) myocardial delineation allows the computation of critical LV functional descriptors (e.g., ejection fraction and wall thickening). The manual contour tracing used in current clinical practice is labor-intensive, time-consuming, and involves considerable inter- and intra-observer variations [1]. The development of computer-assisted myocardial contour extraction will reduce the analysis time and, more importantly, will produce unbiased and consistent results. The seg- mentation of CMR data typically faces two challenges: the determination of the inter-tissue boundaries (blurred due to partial voluming), and the delineation of ⋆ This material is based upon work supported in part by the National Science Fo- undation under Grants IIS-9985482 and IIS-0335578. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. C. Barillot, D.R. Haynor, and P. Hellier (Eds.): MICCAI 2004, LNCS 3216, pp. 402–410, 2004. c  Springer-Verlag Berlin Heidelberg 2004