Improved Mean Shift Segmentation Scheme for Medical Ultrasound Images

Abstract

Medical ultrasound image segmentation is a critical step for image analysis and measurement. In this paper, an improved scheme is proposed to overcome limitations of the traditional mean shift algorithm in medical ultrasound image segmentation application. Two aspects are taken into account to improve the traditional algorithm: adaptive gray scale bandwidth selecting and threshold-based region merging. Experimental results of simulated and clinical ultrasound images indicate that better classification of objects and background could be achieved by selecting gray scale bandwidth adaptively, and computation cost is reduced obviously by employing simple threshold-based region mergence. In addition, small patches associated with ultrasonic speckle textures could be eliminated effectively.