Adaptive Segmentation of Multi-modal 3D Data Using Robust Level Set Techniques Aly Farag and Hossam Hassan Computer Vision and Image Processing Laboratory University of Louisville, Louisville, KY 40292. {hossam,farag}@cvip.uofl.edu http://www.cvip.louisville.edu Abstract. A new 3D segmentation method based on the level set tech- nique is proposed. The main contribution is a robust evolutionary model which requires no fine tuning of parameters. A closed 3D surface prop- agates from an initial position towards the desired region boundaries through an iterative evolution of a specific 4D implicit function. Infor- mation about the regions is involved by estimating, at each iteration, parameters of probability density functions. The method can be applied to different kinds of data, e.g for segmenting anatomical structures in 3D magnetic resonance images and angiography. Experimental results of these two types of data are discussed. 1 Introduction Both surgical planning and navigation benefit from image segmentation. Also the 3D segmentation of anatomical structures is very important for medical visualization and diagnostics. The segmentation process is still a challenging problem because of image noise and inhomogeneities. Therefore this process can not depend only on image information but also has to exploit the prior knowledge of shapes and other properties of the structures to be segmented. In many cases, the 3D segmentation is performed using deformable models. The mathematical foundation of such models represents the confluence of physics and geometry [1]. The latter represents an object shape and the former puts constraints on how the shape may vary over space and time. Deformable models have had great successes in imaging and computer graphics. In particular in [2], the deformable models recover the object’s structure using some properties of its shape. The model evolves iteratively towards the steady state of energy minimization. But the disadvantage of this method is that the initial contour should be close to the final one. The model faces also problems with topological changes of a complex structure. Level set techniques of segmentation overcome problems of the classical de- formable models [3,4,5]. A curve in 2D or a surface in 3D evolves in such a way as to cover a complex shape or structure. Its initialization is either manual or automatic and it need not to be close to the desired solution. But these methods depend on a big number of parameters to be tuned for the success of the process. C. Barillot, D.R. Haynor, and P. Hellier (Eds.): MICCAI 2004, LNCS 3216, pp. 143–150, 2004. c  Springer-Verlag Berlin Heidelberg 2004