Abstract:

A modified normalized cut (Ncut) method considering the weighted gray values of neighborhood pixels is proposed to automatically segment the breast tumor lesion in ultrasonic images. The method partitions a breast ultrasound image into clusters with Ncut, and uses different gray values and the spatial distribution of each cluster to obtain an initial contour of the breast tumor. Then, for a small percentage of inaccurate segmentation, an active contour model together with a region-scalable fitting term is used to adjust the initial boundary for the final result. The proposed method is applied to a database of 112 clinical ultrasonic breast tumor images. The results show that the proposed method can realize boundary extraction of tumors efficiently and automatically without any manual intervene. Meanwhile the computation complexity is low. Therefore, the method can be used to improve the degree of automation in computer-aided diagnosis.

Key words: ultrasound images, breast tumor, boundary, automatic extraction, normalized cut, region-scalable fitting