Research On Breast Tumor Detection Based On Single Voxel Feature Dictionary

Breast cancer is extremely general in female group.In recent years,the incidence of breast cancer has not declined but begins to increase.However,the development of the disease can be effectively contained through early diagnosis and even save a considerable number of patient’s life.Among them,mammography is the most effective method for the early detection of breast cancer.But,early tumors don’t form significant morphological changes such as early superficial lesions which are difficult to screen by morphological testing.Aim at those problems,this thesis intended to study early tumor screening methods based on voxel levels,and carried out the following research:(1)Image preprocessing.we delete various interferential factors unrelated to this experiment such as some marks and backgrounds by preprocessing the image(such as region growing).Then the image is separated.The tumor portion of the image is stored separately from the normal tissue portion,waiting for the next treatment.(2)Establish a dictionary of priori features based on monomeric features.First,wavelet transform and pixel corrosion of the image is performed.Then,we extract the features of each voxel in the segmented breast area of the image,and construct a dictionary of breast tumor prior characteristics and a dictionary of normal features of breast normal tissue.(3)Online search and evaluation based on priori feature dictionary.In this thesis,we introduce the concept of weight to describe the reliability of each prior sample.The initial value of each sample’s weight is assigned to each image segment,and the artificial intelligence learning method is used to adjust the weight of each tumor based on the Euclidean distance according to the prediction function of each segment in the priori feature dictionary,for realizing the automatic correction of the weight score.Then,the image to be recognized is judged pixel by pixel in such a line search manner: if the connected aggregation region of the tumor voxel reaches a certain number,the region is considered to be a suspected tumor region.In the detection experiments,the detection accuracy of the proposed method is 93.3%,which is better than the classical detection method.It can help doctors to detect breast tumors and improve their work efficiency.