Combining Texture Features with Extended Gradient Vector Flow Field for Automatic Detection and Segmentation of Lesions in Ultrasound Liver Images

Abstract

Ultrasound image has the advantage of being non-invasive, externally applied, non-traumatic, but they are usually displayed in gray scale and exhibit many artifacts such as coarse resolution, low contrast and high noise. So interpretation of the ultrasonic image by human eye becomes more difficult and hence error prone. In this paper, we proposed a technique for automatic detection and segmentation of lesions in liver sonograms. For this an algorithm based on the Extended Gradient Vector Flow (E-GVF) field model is proposed for seed selection. The scoring and selection of seeds by considering their local gradient direction information and texture feature information. Then the region growing algorithm is used for the segmentation of lesions. Thus this method performs satisfactorily for the automatic detection and segmentation of single and multiple lesions in ultrasound liver images. The system was validated on a data base of liver sonograms for 20 patients. The true positive area overlap was 76.61%. Running time for segmenting a single sonogram image was 24 s on a 1.4 GHz Celeron processor for the size of 168 X 119.