Study On Pulmonary CT Image Nodule Segmentation And Classification Of Benign And Malignant Based On Radiomics

Studies have shown that lung cancer is one of the most malignant tumors in the world with the highest morbidity and mortality.The clinical manifestations of lung cancer are complicated,and the symptoms of pulmonary nodules vary from person to person,which brings great trouble to the clinical diagnosis and treatment.Radiomics is a research focus in the field of medical imaging.It aims to use computer aided diagnosis(CAD)technology and image processing algorithms to mine quantitative feature data from medical images.Through quantitative analysis of large amounts of data,screening the most valuable data features,and then conducting deeper data mining,prediction and analysis to assist clinicians to obtain accurate diagnostic results.Based on Radiomics,this paper focuses on pulmonary nodule segmentation and benign and malignant classification of pulmonary CT images.The main research work is as follows:(1)Aiming at the problem of pulmonary nodule segmentation,this paper proposes an image segmentation algorithm based on Fuzzy C-Means(FCM)clustering and improved Random Walk(RW).First,using an improved FCM algorithm combined with anisotropic diffusion filtering and morphological processing to record pixel vertex coordinate information on the map and provide seeds for the random walk algorithm;Secondly,the geodesic distance is added to improve the weight function of the random walk algorithm.The improved random walk segmentation algorithm is used to complete the accurate segmentation of pulmonary nodules,which avoids the over segmentation problem caused by the interference of gray level and blood vessels in the traditional segmentation algorithm,shorten the segmentation time of a single pulmonary CT image.(2)Aiming at the benign and malignant classification of pulmonary nodules,this paper first extracts 95 high-throughput Radiomic features of pulmonary nodules regions segmented using the improved random walk algorithm;secondly,an M-LASSO regression model based on Mahalanobis distance and least absolute shrinkage and selection operator(LASSO)is proposed,the optimal feature subset filtered by the model is input to the Support VectorMachine(SVM)classifier to complete the benign and malignant classification of pulmonary nodules,which improves the classification accuracy effectively.(3)This article combines research theory with clinical practice.Based on the pulmonary nodule segmentation and benign and malignant classification algorithms proposed in this paper,Java language and Matlab language are used to design a lung cancer assistant diagnosis platform.The platform runs stably and is easy to operate,it can assist clinicians to analyze and diagnose,improve work efficiency,and reduce the rate of misdiagnosis.