Segmentation of ultrasound imagery using anisotropic diffusion

We present an anisotropic diffusion technique, tailored to ultrasound images, for the purpose of enhancing and segmenting medical ultrasound imagery. Taking into account the speckle characteristics of ultrasound imaging, we introduce a specialized set of partial differential equations to adaptively enhance the raw ultrasound data. The diffusion method mimics traditional anisotropic diffusion in the sense of directional sensitivity. As a result, homogeneous regions are smoothed in an intra-region manner and edges can be enhanced and extracted in the presence of speckle. We provide a series of experiments using synthetically developed models of the carotid artery. The experiments investigate the effect of varying edge orientation, level of noise, and amount of image contrast. Results from the novel diffusion technique are compared with the conventional approaches, revealing the efficacy of the proposed method.