Automating 3D Echocardiographic Image Analysis Gerardo I. Sanchez-Ortiz and J´ erˆome Declerck, Miguel Mulet-Parada, and J. Alison Noble Medical Vision Laboratory, Department of Engineering Science, University of Oxford, Parks Road, Oxford OX1 3PJ, UK. giso@robots.ox.ac.uk http://www.robots.ox.ac.uk/∼mvl Abstract. 3D echocardiography is a recent cardiac imaging method ac- tively developed for quantitative analysis of heart function. A major barrier for its use as a quantitative tool in routine clinical practice is the absence of accurate and robust segmentation and tracking methods necessary to make the analysis fully automatic. In this article we present a fully-automated 3D echocardiographic image processing protocol for assessment of left ventricular (LV) function. We combine global image information provided by a novel multi-scale fuzzy-clustering segmenta- tion algorithm, with local boundaries obtained with phase-based acoustic feature detection. We fit and track the LV surface using a 4D continuous transformation. To our knowledge this is the first report of a completely automated method. The protocol is viable for clinical practice. We ex- hibit and compare qualitative and quantitative results on three 3D image sequences that have been processed manually, in semi-automatic man- ner, and in fully automated fashion. Volume curves are derived and the ejection fractions errors with respect to manual segmentation are shown to be below 5%. 1 Introduction The last few years have seen the emergence of 3D echocardiography acquisition systems in the market. Methods of acquisition are improving (in terms of spatial and temporal resolution), moving now towards real-time volumetric acquisition. However, interpretation and analysis of the data is more complex and time con- suming than for conventional 2D echocardiography. As recent research studies have shown [1,2,3], the use of three-dimensional data provides more precise in- formation on the pathophysiology of the heart than conventional analysis of 2D views ([4,5,6] and references therein), especially for volume and ejection fraction calculation. Previous work has shown the feasibility of reconstructing a three-dimensional surface of the heart from sparse views [2]. However, in that work, the amount of interaction required to obtain a reconstruction of the endocardium was very large. In [1], a 3D finite element mesh of the left ventricular (LV) myocardium is computed and used to perform strain analysis. The approach is interesting, S.L. Delp, A.M. DiGioia, and B. Jaramaz (Eds.): MICCAI 2000, LNCS 1935, pp. 687–696, 2000. c  Springer-Verlag Berlin Heidelberg 2000