Studies Of Ultrasound Tumor Image Segmentation Algorithm

In recent years,despite the people’s living standards are getting higher and higher,but a variety of cancer diseases on human life is still causing great harm.With the continuous progress of science and technology,the people’s requirements of the medical equipment are getting higher and higher.By focusing the ultrasound probe on the tumor tissue,to achieve the treatment of tumor ablation.This treatment program does not require the use of a scalpel to cut the tumor to the patient,but directly through the instrument in the patient outside the tumor ablation,with a high therapeutic efficiency.Ultrasound tumor image segmentation task mainly by the professional physician to manually complete,which gives professional physicians caused heavy task.Therefore,in the process of tumor image segmentation to minimize the intervention of the human body,the use of computers to segment the tumor image,this study for HIFU treatment has a high value.Since the ultrasound image has the following characteristics:a.Low resolution,low signal to noise ratio;b.Low contrast,weak edge;c.Acoustic artifacts and speckle noise.Because of these features,the separation of ultrasound images is a very difficult task.In view of these problems,we must first pretreatment of ultrasound images.In this paper,we study the related preprocessing methods.In this paper,three methods for segmenting ultrasound images are discussed.One is the image segmentation algorithm based on the idea of splitting and merging.The splitting step in this algorithm uses the watershed transform to construct the superpixel.The merging step takes the ultrasound image over-segmented area as the superpixel and then uses the MDL principle to merge,which will have similar texture features small areas clustered together,resulting in tumor segmentation images,splitting and merging method to overcome the problem of poor quality of ultrasound images.The second is based on the gradient threshold of the watershed transformation,by improving the seed point selection method to solve the problem of watershed over-segmentation,directly used to achieve tumor image segmentation.The third is the ISODATA clustering method.By analyzing the position information and gray level information of the image as the characteristics to be classified,the optimal number of clusters is obtained by multiple iterations to realize the segmentation of the ultrasound tumor image.