Research On Ultrasound Image Segmentation Based On Texture Position Constraint

The number of cancers in China is increasing,and like many other countries around the world,the cancer is becoming one of the important threats to people’s health.and,the breast cancer is one of the cancers with high incidence.Clinically,if the breast cancer can be found and treated in early,the probability of recovery will be greatly improved.With ultrasound diagnosis,according to different tissue echo intensity and different attenuation degree,the accuracy of screening for breast cancer can be greatly increased.Because of the high incidence of different age groups and regions is different,the number of screenings received by hospitals is gradually increasing.Long-term artificial judgment is easy to cause errors in doctor fatigue and diagnosis.If computer aided systems can be used,automated help description doctors complete certain initial screening tasks,and the diagnostic pressure on doctors and hospitals will be greatly reduced.In the field of ultrasonic medical image processing,it is often first to segment the target image and then identify it.The main research direction of this paper is to target the ultrasound image,which is the method of medical image semantic segmentation.The semantic segmentation method of medical images is to classify pixels or pixel blocks of the target image to determine the category of each pixel or pixel block to achieve the purpose of segmenting the image.In this paper,we use the spatial context feature to study the segmentation method of ultrasound medical images in order to obtain better segmentation effect.Our paper first analyzes and summarizes the research status of image segmentation at home and abroad,and then selects and uses the widely used methods in medical image field to analyze and summarize them.Next,the idea of the segmentation method of this study is proposed.Because of the serious noise,low contrast and unclear edge of the ultrasound image,it is less effective to segment the image by relying only on the gray information.Our study introduces Gabor texture feature information to compensate for the lack of gray information and improve the segmentation effect.Because the ultrasound breast image has strong structural information,this study extracts the direction-aware spatial context feature(DSC)and uses the idea of RNN network to improve the segmentation model’s use of location information.Our experiments show that the proposed segmentation model based on texture information and position information has achieved certain results in the segmentation of ultrasound mammography images.