Contour extraction from ultrasound images viewed as a tracking problem

This paper focuses on the problem of contour extraction in medical images and shows that this segmentation task can be viewed as a tracking problem. The influence of the speckle noise and of the high non-linearities is overcome by Monte Carlo methods. An efficient multiple model Monte Carlo algorithm for progressive contour growing (tracking) is developed, accounting for convex, non-circular forms of delineated contour areas. The driving idea of the proposed filter consists in the incorporation of different image intensity inside and outside the contour into the filter likelihood. A PDA procedure using the contour edge magnitude is applied to cope with the measurement origin uncertainty. The filter performance is studied by extracting contours from a number of real and simulated ultrasound medical images and very good accuracy is achieved.