Segmentation of Breast Nodules on Ultrasonographic Images Based on Marke d-Controlled Watershed Transform

in this article is presented a computerized segmentation method for breast nodules on ultrasonic images. with the goal of removing the speckle while preserving important information from the lesion boundaries, a gabor filter followed by an anisotropic diffusion filtering are applied to the ultrasonic image. furthermore, the marker-controlled watershed transform defines potential boundaries that maximize the average radial derivative function to get the final lesion contour. the segmentation procedure was applied on a database of 50 images and the computer-delineated margins were compared against manual outlines drawn by two radiologist. this comparison was performed by two metrics, which measure the similarity between two compared images: overlap ratio (or) and normalized residual value (nrv). if there is perfect agreement between both images or = 1 and nrv = 0. then, the mean values results, for each metric, were for the first radiologist: or = 0.87±0.04 and nrv = 0.14±0.06, and for the second radiologist: or = 0.86±0.06 and nrv = 0.15±0.05.