International Congress Series 1281 (2005) 1274
Bounding ellipse fitting to head images for accurate active contour initialization K. PataridisT, S.J. Golding, I. Evangelou Department of Radiology, University of Oxford, United Kingdom
Keywords: Image processing; Head MRI; Active contours
1. Introduction In this paper, we report a method developed for fitting an ellipse onto the contour of the frontal and parietal bones of the skull, in sagittal Magnetic Resonance (MR) images of the head. The derived ellipse functions as a bounding polygon for the enclosed soft tissue. Points that lie on this polygon can be used as initialization contour points for an active contour model looking into segmentation of the brain. 2. Methods The algorithm receives the MR image as input, and produces the bounding polygon as a sequence of 2D vertices. It is split into the following distinctive tasks:
! Application of a bilateral filtering to remove noise and smaller details in soft tissue while retaining the details of region boundaries;
! Use of the Canny edge detection algorithm on the image to produce continuous binary edges. ! Edge tracking of contours, based on local continuity of edge strength and direction. ! Encoding of contours into best-fit ellipses. The ellipse with the maximum coverage area is selected as a representation of the estimated bounding ellipse of the structure defined by the upper body of the skull. The initial points of a dynamic contour model can then be set as those of an equilateral n-gon that lies on the derived ellipse. 3. Results The algorithm performs with satisfactory results under different MRI sequence setups. The contour of the upper part of the skull is always present and continuous. In ellipse fitting, the mean square error between the ellipse and the underlying contour is low (signifying that the shape of the bone structures in question is indeed in the form of an ellipsoid). With regard to choosing, out of the number of different ellipses that are estimated, the one that bounds the brain, this is a very straightforward affair, with small capacity for error: the selected ellipse has an area significantly greater than any other. 4. Conclusions The ellipse fitting model can be used as a simple way of determining the initial conditions for an active contour. It has been tested with several MR images of the head. However, the model can be used to bind any anatomical feature that is surrounded (partially or totally) by relative high-contrast structures. Additionally, the model could act as a landmark coordinate system for these anatomical features. T Corresponding author. E-mail address:
[email protected] (K. Pataridis). 0531-5131/ D 2005 CARS & Elsevier B.V. All rights reserved. doi:10.1016/j.ics.2005.03.308