A novel breast ultrasound image automated segmentation algorithm based on seeded region growing integrating gradual equipartition threshold

Abstract

Automatic breast ultrasound (BUS) lesions segmentation based on seeded region growing (SRG) algorithm needs to solve two critical procedures: automatic selection of seed points and the segmentation threshold without manual intervention. For the former procedure, we establish two constraints combining iterative quadtree decomposition (QTD) and the gray characteristics of the lesion to locate the seed inside the lesion. For the latter procedure, the gradual equipartition algorithm according to the maximum change rate of the extracted region is adopted to take infinite approximation to the optimal threshold. The method is testified with 96 BUS lesion images. Quantitative results demonstrate that the proposed method can automatically find out the seed within the lesion with an accuracy rate of 92.27%. More importantly the average time consumed by the proposed algorithm is 12.02 s. Under the condition of large image samples, the efficiency is higher than that of manual segmentation.