Research On The Algorithm For Diagnosis Of Pulmonary Nodules In CT Imaging On Support Vector Machines And Radom Forests

Lung cancer is a malignant tumor with high mortality and high incidence,which seriously threatens the safety of human life and property.Early detection and diagnosis of lung cancer can improve patient survival.Due to the huge volume of CT images,manual review is prone to misdiagnosis and missed diagnosis.Therefore,it is necessary to assist doctors to complete the diagnosis of pulmonary nodules with the help of machine learning,that is,the intelligent auxiliary diagnosis system for pulmonary nodules.However,this auxiliary diagnosis technology has some problems,such as difficulty in accurate segmentation of target regions and low classification accuracy.In response to the problems raised,this paper proposes two methods for diagnosing pulmonary nodules.The main research contents are as follows:(1)We propose a pulmonary nodule diagnosis method based on P_LDA and support vector machine.This algorithm introduces mean clustering into image preprocessing to achieve the purpose of distinguishing blood vessels and pulmonary nodules.The algorithm also combines adaptive thresholding,hole filling and morphological opening and closing operatioms to obtain regions of interest.Considering that the extracted mixed features contain a lot of redundant noise,this algorithm uses the P_LDA method to select high-dimensional data.In addition,the algorithm uses the model constructed by GA-SVM for nodule detection.Extensive experiments show that the diagnostic system combining P_LDA and GA-SVM has high detection and classification accuracy of nodules.(2)We proposed improved anisotropic and random forests based benign and malignant pulmonary nodules a diagnosis method.Considering the lots of noise in lung CT images,which affects the diagnostic accuracy of the system,the P-M model optimized by the piecewise diffusion coefficient function is used to preprocess the images.Aiming at the under-segmentation and over-segmentation of existing problems,an algorithm for improving geometric contour segmentation is proposed by reconstructing the energy function.For the benign and malignant pulmonary nodules of classification,a random forest classification model was constructed.A large number of experiments show that it has obvious advantages in image denoising and lung nodule segmentation.The random forest classifier based on the extracted features by the model is betterr than KNN,BP and SVM in the diagnosis of benign and malignant nodules.