Contour tting using stochastic and probabilistic relaxation for Cine MR Images

Image segmentation by energy{minimizing active contour models (snakes) su ers from the fact that classic numerical optimization algorithms nd only local energy minima, which makes the approach very sensible to noise and initialization. This paper presents a robust adaptive snake model using a stochastic relaxation technique, Simulated Annealing (SA), to nd a global energy minimum in noisy cine MR images. Once a reliable segmentation has been done, a much faster probabilistic relaxation technique, Iterated Conditional Modes (ICM), is used for energy minimization in the following time frames.