Multiresolution Active Contour Model: Application to Lung and Colon Images

J. Dehmeshki, I.C. Ster, M.M. Siddique, X. Ye, M.V. Casique, and X.Y. Lin (UK)


Wavelet Transform, Multiresolution, Snakes


This paper deploys a wavelet based scale-space approach to extract the boundary of the object of interest in medical CT images. The classical approach of the active contour models consists of starting with an initial contour, to deform it under the action of some forces attracting the contour towards the edges by means of a set of forces. The mathematical model involves in the minimization of an objective function called energy functional, which depends on the geometry of the contour as well as of the image characteristics. Various strategies could be used for the formulation of the energy functional and its optimization. In this study, a wavelet based scale-space approach has been adopted. The coarsest scale is able to enlarge the capture region surrounding an object and avoids the trapping of contour into weak edges. The finer scales are used to refine the contour as close as possible to the boundary of the object. An adaptive scale coefficient for the balloon energy has been introduced. Four levels of resolution have been applied in order to get reproducibility of the contour despite of poor different initializations. The scheme has been applied to segment the regions of interest in CT lung and colon images. The result has been shown to be accurate and reproducible for the cases having nodules containing fat, holes and other small high intensity objects inside as well as for colon polyps.

Important Links:

Go Back