科研之友
微信
新浪微博
Facebook
分享链接
摘要
Automatic segmentation of breast lesions in 2D ultrasound B-scan images via active contours, require a seed point to be selected inside the breast lesion. The grey levels on an ultrasound image of the breast show intensity information. The fat tissue is hypo echoic relative to the surrounding glandular tissue. The glandular parenchyma tissue usually appears homogeneously echogenic as compared with fat lobules. Simple cysts are anechoic. Malignant solid masses are usually heterogeneous, hypo echoic and tend to look intensely black compared to surrounding isoechoic fat. Benign solid masses tend to appear on ultrasound with intense and uniform hyper echogenicity. Texture features represent changes in grey level intensities. This paper proposes a method that can automatically identify a seed point based on texture features and allow automatic contour initialization for level set segmentation. This seed point plotted on an US B-scan image is mapped on to its corresponding elastogram pair. The proposed approach is applied to 199 ultrasound B-scan images of which 52 are benign solid masses, 84 malignant solid masses and 63 simple and complex cysts. The seed point obtained using this approach is mapped to its corresponding elastogram pair in 62 US B-scan and US elastography image pairs. Quantitative experiment results show that our proposed approach can successfully find proper seed points based on texture values, in ultrasound B-scan images and therefore in elastography images, with an overall accuracy of 86.93%. This approach is effective and makes segmentation of breast lesions computationally easier, more accurate and fast.
- 出版日期2015
