Anisotropic Diffusion for Medical Image Enhancement

Advances in digital imaging techniques have made possible the acquisition of large volumes of Trans-rectal Ultrasound (TRUS) prostate images so that there is considerable demand for automated segmentation of these images. Prostate cancer diagnosis and treatment rely on segmentation of TRUS prostate images. This is a challenging and difficult task due to weak prostate boundaries, speckle noise, and narrow range of gray levels which leads most image segmentation methods to perform poorly. Although the enhancement of ultrasound images is difficult, prostate segmentation can be potentially improved by enhancement of the contrast of TRUS images. Anisotropic diffusion has been used for image analysis based on selective smoothness or enhancement of local features such as region boundaries. In its conventional form, anisotropic diffusion tends to encourage within-region smoothness and avoid diffusion across different regions. In this paper we extend the anisotropic diffusion to multiple directions such that segmentation methods can effectively be applied based on rich extracted features. A preliminary segmentation method based on extended diffusion is proposed. Finally an adaptive anisotropic diffusion is introduced based on image statistics.