A hybrid approach of tumor segmentation in ultrasound images based on Chan-Vese model and information theorem

An effective hybrid approach that integrates active contour model of Chan-Vese and information theorem for tumors segmentation in medical ultrasound (US) images is proposed in this paper. Our model can detect objects whose boundaries are not necessarily defined by gradient. In addition, taking advantages of the local statistical information of regions, mutual information is added to Chan-Vese model to form the total energy function for tumors segmentation task in US images. Comparing with Chan-Vese model and level sets method, the proposed hybrid approach can characterize the local features of a tumor in US images more efficiently. Experimental results have shown that our method could achieve higher segmentation accuracy in comparison with the other two widely used segmentation methods, and be capable of segmenting tumors in US images.