Robust B-spline snakes for ultrasound image segmentation

Segmenting echographic images is a difficult task due to their low contrast and to the presence of speckle. This paper presents an original robust B-Spline snake based on a novel external energy. Our approach combines anisotropic diffusion with the process of curve evolution by using a local coefficient of variation and the Tukey's error norm. This makes the snake robust to speckle. At each iteration the algorithm computes an edge image, and derives a new external energy based on the amplitude and direction of the gradient of the upper mentioned coefficient. The method has been tested on echocardographic images of a 12-week old foetus. The results, presented with different parameters, show a significant improvement.