A novel region growing method for segmenting ultrasound images

Abstract

Region growing is performed in a multi-feature vector space. Three novel criteria are developed for region growing control. First, instead of using the local information as do the conventional region growing methods, this method uses global information. Second, to overcome the effects of speckle noise and attenuation artifacts, a new idea termed, "geographic similarity", is introduced. Third, an equal opportunity competence criterion is employed to make results independent of processing order. Segmentation results for in vivo intracardiac ultrasound images and the corresponding statistical analyses show that this method is reliable and effective.