Research On Classification Of Multi Kernel SVM Pulmonary Nodules Based On Fitness Feedback PSO Optimization

Since 21-Century,the incidence and fatality rate of lung cancer have been improved in various parts of the world.Especially in the period of rapid development of China,the rapid industrial development in recent years cause environmental degradation,air pollution and other issues,the number of lung cancer infections and deaths in China are ranked the first number in the world.The initial characterization of lung cancer is not obvious,delay the best treatment period,cause lung cancer five years survival rate is still lower than the average level of cancers.According to the American Cancer Association survey,the survival rate of patients with early diagnosis of benign and malignant pulmonary nodules is as high as 80%,while the survival rate of patients with middle and late lung cancer called III and IV times are not more than 20%obviously.It can be seen that early detection of pulmonary nodules and early diagnosis of benign and malignant are important means to reduce the mortality of lung cancer.With the development of medical imaging technology,compared with invasive biopsy methods,classification of benign and malignant pulmonary nodules based on medical images has become a hot topic in the field of medical image processing.A variety of benign and malignant diagnosis methods of pulmonary nodules based on image data have been proposed.This article is based on the pulmonary nodule image data to study the following three aspects.1)Due to the heterogeneity of pulmonary nodules,the features extracted by conventional feature extraction methods contain redundant features with repetitive information.Based on the detailed study of the PCA method based on dimensionality reduction,a feature extraction and feature selection method using PCA for pulmonary nodules is proposed.Finally,12 sets of optimal feature sets for building and validating classification models are determined.After that,the extracted 12 sets of features are processed by 0 means normalization,which are used as training samples and verification samples.2)With the detailed analysis of various types of benign and malignant pulmonary nodules classification method and the shortcomings of the single kernel function in heterogeneous data.As well as the low accuracy and poor generalization capability of the most of benign or malignant pulmonary nodules classifier,and combine with the series of advantages of the multi-kernel SVM method in dealing with the heterogeneous data and the complex samples.A multi-kernel SVM method is proposed for the classification of pulmonary nodules.3)In the light of the standard PSO algorithm is easy to be fitted to local optimum in the process of multi-kernel function parameter optimization,and the particle oscillation is bigger in the late iteration stage.PSO algorithm based on fitness feedback effect is proposed to optimize the kernel parameters of multi-kernel SVM.In this experiment,90 cases of lung CT data were used from Zhejiang Jinhua Guangfu Tumor Hospital for nearly three years.The accuracy and AUC values of the three kernel functions are compared objectively by experiments.The results show that the classification accuracy of the training samples and the validation samples in this method is as high as 91.47%and 88.61%respectively,and AUC=0.899.Through experimental verification,contrast on single kernel SVM method,the classifier constructed based on multi kernel SVM method has been improved in generalization capability and learning capability,and it also improve classification accuracy effectively.The experimental results indicate that this method is viable in the clinical diagnosis and treatment of lung nodules,and can be used as a non-invasive method to assist doctors in the early prediction of pulmonary nodules.