Robust Segmentation For Breast Ultrasound Image Based On Saliency Detection

Breast cancer is the highest incidence of cancer in female patients.Breast cancer has caused great harm to women's health and also caused great distress to their psychological and life.Clinical studies have shown that the early detection of breast cancer and effective treatment can greatly improve the survival rate of female patients.Ultrasound imaging technology of medical imaging,with low cost,good diagnostic effect,good image quality,no harm to the human body and flexible and so on,has become a routine examination in the diagnosis of breast.Computer-aided diagnosis(CAD)is widely used in the diagnosis of breast cancer because it can help doctors to diagnose,improve the accuracy and objectivity of diagnosis,reduce the errors in the diagnosis process and improve the efficiency.The location of the lesion in the computer-aided diagnosis directly affects the qualitative and quantitative analysis of the follow-up.The image segmentation of the lesion becomes the most important part of the computer-aided diagnosis.Most of current tumor segmentation method is not fully automatic,these methods need doctors to select seed points,draw the region of interest,etc.When dealing with large amounts of cases,it can burden doctors and slow down their work,making it easier for doctors to make the wrong judgments.Ultrasound images of breast can be collected through a variety of different devices.There are relatively large differences in the quality of images collected by these devices,such as low resolution,loss of detail,low contrast and so on.The current segmentation scheme is often based on the image acquired by a certain device.Owing to the variety of devices,the segmentation method does not apply to all images.To solve the problem of breast segmentation mentioned above,we proposed a fully automatic segmentation method based on visual saliency in this paper.We found that breast images collected by different devices remained unchanged in the anatomical structure of the breast.Therefore,we combine this feature with the feature of ultrasonic images to construct saliency maps based on boundary cues and background contrast cues.The two saliency map represent two di fferent characteristics of the breast ultrasound images,and the first stage saliency map is obtained by fusing them.Due to the influence of the background contrast cues,there is an internal inhomogeneity in the first stage of the saliency map.And the f inal saliency map is obtained by selecting foreground seed point and manifold ranking.The experimental results show that our method can not only effectively eliminate the influence of hypoechoic background,but also can locate the tumor well.