Research Of Ultrasound Image Classification Algorithm Based On Semantic Features

We know that the incidence rate of thyroid nodules has been increasing recently years.As thyroid cancer symptoms are not obviously and the subjective experience of doctor, thereis a big deviation diagnosis. With the rapid development of modern technology of computerand medical, now the application of computer aided diagnosis has been widely used in ourdaily life.In this paper, the main object of study is the thyroid ultrasound image. Because the limitof characteristics of benign and malignant of thyroid nodules are not clear, we can notrecognize them, which is similar to the ambiguity and synonyms in text. The medical imagescontains lots of implied information and specific structure. We can not solve the underlyingcharacteristics of traditional until the presentation of PLSA algorithm. PLSA could solve theproblem of ambiguity and synonyms in text. So PLSA algorithm is applied to the processingof thyroid ultrasound image. By extracting the semantic features of image, we can distinguishbetween benign and malignant.Support vector machine(SVM) can solve small sample and nonlinear problems, that issuitable to classify image features, so we introduce it into the classification of thyroidultrasound image and then input SVM by using the extractive semantic features todifferentiate.The main content of the paper is as follow:1、In order to denoise and enhance the ultrasonic image by using anisotropic diffusionmodel and then improve the image clarity. We can make use of SIFT feature descriptors toextract image feature points and form feature points library;2、 Cluster feature points Through the K-means algorithm, so we can remove itsredundancy and form visual word; Basing on counting the frequency number of each visualword in each image and forming Visual words and images in the matrix. Then you need tosend them to a PLSAto find semantic features.3、 During the extraction of semantic features, the paper studied the influence of classification accuracy according to the number of visual words and different topics; For thepurpose of eliminate drawback of standard SVM, to improve the algorithm by adding similarweighting factor. This method could solve classification of deviation due to a seriousimbalance about benign and malignant thyroid data; Finally, the similar weighted SVMalgorithm and standard SVM algorithm were compared through different experimentation.The results of experiments show that extraction of semantic features can distinguishbetween thyroid nodules’benign and malignant very well; In addition, the SVM algorithm ofadding similar weighting factor has higher classification accuracy,when it is used in theclassification of thyroid nodules’benign and malignant.