C. Barillot, D.R. Haynor, and P. Hellier (Eds.): MICCAI 2004, LNCS 3217, pp. 1052–1054, 2004. © Springer-Verlag Berlin Heidelberg 2004 Construction of a 3D Volumetric Probabilistic Model of the Mouse Kidney from MRI Hirohito Okuda 1,2,3 , Pavel Shkarin 3 , Kevin Behar 4 , James S. Duncan 2,3 , and Xenophon Papademetris 2,3 1 Production Engineering and Research Lab., Hitachi Ltd., Kanagawa, Japan, 2 Department of Biomed. Engineering, 3 Diag. Radiology and 4 Phsychiatry Yale University New Haven, CT 06520-8042 Abstract. We present the results of constructing a probabilistic volumetric model of 3D MR kidney images. The ultimate goal of this work is the mouse kidney segmentation based on a probabilistic volumetric model. The kidneys were aligned into the base shape using an extended robust point matching algorithm. The registration step consists of the global linear transformation and the local B-spline based free form deformation. Shape modeling is performed with globally aligned shape and template volumetric image is generated with locally aligned images. We are currently working on developing a segmentation algorithm using our model. 1 Introduction The ultimate goal of this work is to automate the segmentation of kidneys, and to quantify kidney volume in transgenic mouse models [1] of polycystic kidney disease. Toward this goal, here we present the result of constructing a probabilistic volumetric model which is the first key component of our strategy for the segmentation process. A general drawback of the model constructing methods proposed so far [2,5] is the correspondence problem where the definition of one-to-one mapping across data are needed. To solve this problem, we apply the extended robust point matching algorithm (RPM)[3] which can automatically compute the correspondences. Here we present a result of probabilistic volumetric model of both kidneys constructed using RPM . 2 Method and Results Figure 1(a) shows an example of training image data. Ten postmortem eight-week old C57BL6 wild type mice were scanned. All imaging was performed on a Bruker 4.0T/40 cm bore animal system using a T2-weighted 3D Multi-spin multi-echo sequence (MSME), with a TE=15ms,FOV=4x2.5x1.8cm and an imaging matrix of 256x128x64. An expert user performed the original surface extraction of both kidneys from the images with a software platform originally designed for segmenting the left ventricle