Research On Automatic Segmentation Algorithm Of Adhesive Pulmonary Nodules

Lung cancer is one of the most serious malignancies that threaten human health and life.It has attracted worldwide attention.According to the American Cancer Society,lung cancer has become the leading cause of death in cancer patients.Early detection and treatment of lung cancer is the key to improving the survival rate of patients.How to effectively detect lung cancer has become a topic of universal concern.The early manifestation of lung cancer is pulmonary nodules,so the correct detection and identification of pulmonary nodules is essential.Computer-Aided Detection(CAD)system can effectively help doctors perform early detection and characterization of lung cancer,avoid missed detection,and improve the detection accuracy.Candidate nodule segmentation and identification are the two main components of the lung CAD detection system.For these two parts,the main research content of the paper is:Multiple types of candidate regions(Region of Interests,ROIs)segmentation.In the image preprocessing stage,the image is first denoised using the median filter;then the lung parenchyma is extracted using the Otsu algorithm and the mathematical morphology method,and the contour of the lung parenchyma is repaired using the rolling ball method to avoid incomplete lung contours.And missed nodule of adhesion lung wall type.In the segmentation phase of candidate nodule ROIs,a fast fuzzy C-means clustering algorithm was used to segment candidate nodule ROIs.This method can segment both solitary nodules and adherent pulmonary nodular nodules,and solves the general algorithm only.For a single type of nodule segmentation is effective,the algorithm is generally not applicable to the problem,can meet the requirements of lung nodule segmentation.Lung nodules were identified using cost-sensitive support vector machines.For the extracted sample nodule ROIs,the number of positive and negative samples is unbalanced,and the parameter search of the classifier is easy to fall into the local optimization problem during the training process.The particle swarm optimization algorithm is combined with the cost-sensitive support vector machine algorithm.A cost-sensitive SVM algorithm based on particle swarm optimization was proposed to identify nodules.The experimental results show that the accuracy rate reaches 91.11%,the sensitivity reaches 85.71%,and the specificity reaches 93.55%.Compared with the cost-sensitive support vector machinebased on genetic algorithm,cost-sensitive support vector machine based on grid search method and support vector machine based on particle swarm optimization algorithm,the proposed PSO-CSVM algorithm is superior to the other three.The algorithm can improve the accuracy of recognition when the number of positive and negative samples is not balanced.