Segmentation Of Pulmonary Nodules Using Improved Active Contour Model

Lung cancer is one of the most common malignancies and has caused serious threat to human physical and mental health.Furthermore,it is almost asymptomatic in the early stages,and always detected in middle-late stages when the treatment effect of present methods are unsatisfactory.Therefore,early diagnosis and effective treatment can efficiently improve the cure and survival rate,reduce the physical and psychological pain of patients.The major sign of early developing lung cancer(pulmonary nodules)can be observed by lung CT-scan images,which is one of the most commonly used methods for routine examination of lung cancer.And this method is simple,inexpensive and less damaging to patients.The changes of the nodule's geometry,multiplication rate and surface smoothness are the important parameters for the physician to analyze and judge lung cancer.The accuracy of pulmonary nodule segmentation determines the accuracy of these indexes.Therefore,the accurate segmentation of pulmonary nodules is the core technology to identify and diagnose the lung cancer that is important to analyze the patient's condition and design treatment options for physicians.The segmentation of pulmonary nodules in Computer-Aided Diagnosis(CAD)of pulmonary nodules is the main focus of this paper.The main objective is to accurately segment the major sign of pulmonary nodules in the early stages of lung cancer in order to provide an intuitive and reliable information for screening of early lung cancer.In this paper,we propose a composite model based on the Edge-based Active Contour Model(EACM)combined with improved local entropy activity profile model(RCV,Renyi Chan-Vese)to accurately segment the lung nodules.Firstly,the suspect pulmonary nodules are filtered,denoised and enhanced.After preprocessing,the ROI(Region of Interest)region of suspect pulmonary nodules is obtained by rough segmentation in the lung parenchyma using the characteristics of edge-based active contour model(EACM).The developed method effectively avoids the shortcomings of the traditional manually segmentation methods.Additionally,the obtained ROI is much smaller than the lung parenchyma area,the boundary is close to the edge of the real pulmonary nodules,and the area overlap rate is also controlled within the effective range.This method can reduce the area of fine segmentation and the following segmentation time.Then,the fine segmentation of lung nodules is achieved by using the RCV(Local Entropy-based Active Contour Model).The ROI obtained automatically from the previous stage is used as the initial contour of the RCV model,which can avoid the trouble of setting the initial contour manually.And the gray scale distribution statistics(local entropy)of the external region are used to automatically adjust the weight parameters of the internal and external energy terms in the energy model,which can increase the sensitivity and robustness of the algorithm and make the final segmentation result converge to the nodule's real boundaries as much as possible to improve segmentation accuracy.At last,270 lung CT images provided by the hospital and the Lung Imaging Database Consortium(LIDC)are used for database in our paper to analyze and compare the results of the algorithm with those of the physician manually drawing the region Gap using six evaluation criteria based on the area and the two measurement values based on the distance.Additionally,the CV model,EACM model and EACM-CV model are also used to carry out the segmentation experiment in the same database in order to compare with our method.The experimental shows that our algorithm can achieve better accuracy and robustness in segmentation,and the segmentation result is closer to the golden standard of hand-painting by doctors.Finally,the design of subsequent benign and malignant diagnosis and treatment of pulmonary nodules has a significant reference value.