Ultrasound image enhancement for detection of contours using speckle-reducing anisotropic diffusion

Ultrasound imaging is a method of obtaining images from inside the human body through the use of high frequency sound waves. Ultrasound imaging suffers from severe speckle contamination that renders the human visualization of ultrasonic imagery and the detection of boundaries of different tissues in ultrasonic images difficult. Edge-preserving speckle filtering methods have been intensely studied to improve the quality of speckle contaminated imagery. Such filters have been shown to be capable of improving the image quality in terms of visualization. However, with these filters, the speckle level remains high in a relatively large neighborhood of an edge after filtering. As a result, these filters do not serve as effective precursors to edge detection. For optical imagery, anisotropic diffusion has been applied to remove noise, while, at the same time, enhancing image edges. When applied directly to ultrasound imagery, anisotropic diffusion fails to remove noise and enhance edges, as it is based on an additive noise model. In this dissertation, we derive a novel diffusion model called speckle reducing anisotropic diffusion (SRAD) for filtering ultrasound imagery while enhancing edges. The novelty of SRAD lies in replacing the gradient magnitude operator in the anisotropic diffusion model by a derived instantaneous coefficient of variation operator, which is formulated as a function of a normalized gradient operator and a normalized Laplacian operator. The dissertation analyzes the edge detection and localization performance of the instantaneous coefficient of variation when posed as an independent edge detection problem. Analysis and testing shows that the proposed SRAD not only improves the quality of imagery for human visualization but also improves edge detection. The SRAD and the instantaneous coefficient of variation are used to construct an active contour algorithm for the segmentation of prostate from transabdominal US imagery. Experimental results are shown and compared with manual outlining of the prostate. SRAD and the instantaneous coefficient of variation show potential for application to quantitative, diagnostic ultrasound.