Research On Kernel Method Based Diagnosis Of Breast Nodules

Breast cancer is a type of cancer that is often found in women and is becoming the main cause to lead in the world.However,people do not fully understand the cause of breast cancer at present.Early detection and treatment can improve the survival rate of this disease and has far-reaching significance.The malignant breast nodules is also known as breast cancer.Whether the benign nodules or malignant nodules,which will bring threat for the patient's physical and mental health.Accurate identification of benign and malignant breast nodules is essential for the next treatment.Therefore,how to use computer-assisted diagnosis and treatment technology as much as possible to assist doctors to diagnose the disease is the purpose of this study.In the 1990 s,the kernel function method(referred to as the kernel method)in the pattern recognition and machine learning has attracted wide attention.As the theoretical development and mature of the kernel function,many of the algorithms that perform well are related to the addition of kernel functions.Therefore,this paper mainly uses the kernel-based algorithm to assist in the diagnosis of benign and malignant breast nodules.Breast nodule ultrasound images are high-dimensional image data,and if we do not do any treatment and train classifier directly which will lead to dimension disaster and inefficiency and other issues.Therefore,in order to improve the classification accuracy and classification efficiency,it is necessary to carry out the dimension reduction of the original high dimensional ultrasonic image.The main work of this paper is as follows:1)This paper studies some classical reduction dimensionality methods including principal component analysis(PCA),linear discriminant analysis(LDA),kernel principal component analysis(KPCA)and Gaussian process latent variable model(GP-LVM).The dissertation uses PCA,KPCA and GP-LVM do reduction dimensionality experiment for breast nodules picture.Through detailed theory analysis and experiment comparison,the dissertation selects the reduction dimensionality results of GP-LVM as input data of classifier finally.2)The image data through GP-LVM are randomly sampled with return five times of this dissertation.It obtains model average value of five group different training sets and testing sets.The dissertation proposes three different classification methods to construct classification model which are K-Nearest Neighbor(KNN),Decision Tree(DST)and Support Vector Machine(SVM).It studies the applicability,advantages and disadvantages of these three methods.After several experiments,the dissertation compares average classification accuracy of these three classifiers.Finally,we can get the classification performance of the SVM with Gaussian radial basis function is generally superior to other classifiers.3)Finally,in order to verify the necessity of reduction dimensionality,this dissertation classifies the data without dimension reduction directly.At the same time,in order to further study the influence of the selection of kernel function for the algorithm,this dissertation chooses different common kernel functions to GP-LVM and SVM to handle breast nodules image.