Abstract:

Aiming at the problem of  thyroid nodule segmentation of ultrasound images, we proposed the new segmentation algorithm combining kernel fuzzy Cmeans (KFCM) clustering with    improved distance regularized level set evdution (DRLSE) model. The algorithm solved the problem that the DRLSE model was sensitive to initial contour, and the evolution parameters needed to be manually set, and the ability of segmentation for weak edges was poor. Firstly, the KFCM clustering algorithm was used to segment the nodule region coarsely which was regarded as the initial evolution contour after binaryzation. Secondly, the evolution parameters of the level set were calculated automatically using the region surrounded by the initial evolution outline. Finally,  the region of thyroid nodules was segmented by Gaussian regularization rule evolution level set. Comparative experimental results show that the proposed algorithm can improve segmentation accuracy of thyroid nodules, and the nodule area can be segmented accurately even in the presence of noise.

Key words: thyroid nodule, kernel fuzzy C-means clustering, level set, Gaussian regularization