Segmentation of salivary gland tumors in ultrasonic images based on anisotropic diffusion and random walk

Abstract

Aimed at the difficult segmentation of ultrasonic tumor image with strong speckle noise, low contrast and weak boundaries, a novel method for segmentation of ultrasonic image is proposed. In order to suppress speckle noise and enhance the edge details, the anisotropic diffusion algorithm combined with the Laplacian operator is introduced into ultrasound images, of which the operator is able to discriminate the gray changes caused by noise or the edge. Then the random walk model in graph theory is employed to achieve an effective segmentation. Lots of clinical ultrasound images of salivary gland tumor are tested and the experiment results demonstrate that the proposed method possesses the nice properties of anisotropic diffusion algorithm and random walk algorithm, overcoming prone over-segmentation or under-segmentation in traditional random walk. In addition, the method bears a high calculating speed and segments tumor accurately and effectively.