A Study Of Lung Parenchyma And ROI Segmentation Algorithms Based On Computed Tomography Images

With the improvement of people’s living standards.lung cancer has become a threat to humans’ health and one of the leading causes of canccr mortality.Therefore,early detection of lung cancer is conducive to improve the survival rate of patients with lung cancer.However,due to the large amount of medical CT image data,if we only rely on the doctor’s subjective diagnosis,coupled with long hours of work caused by fatigue or other factors.it will often lead to misdiagnosis.Therefore,we need the method of automatic detection for pulmonary diseases to improve the diagnosis speed and accuracy of the doctors and to reduce the intensity of doctors’ work.Thus,accurate lung parenchyma and pulmonary ROI segmentation are the prerequisite for subsequent processing.The accurate segmentation fo the lung parenchyma is the premise for the CT image analysis and the detection of the lung diseases.Therefore.the accurate segmentation method for lung parenchyma is proposed firstly.We crudely separate lung parenchyma and other tissue structures with the method of acdaptive Otsu hinaryzation.Then we use the mask image to extract pulmonary parenchyma roughly from lung CT images.Next,3D region growing method remove lung airway tree which is similar to the gray value of lung region from the initial lung parenchyma.Adaptive threshold selection can effectively avoid the phenomenon of over-segmentation.Furthermore.we use line scan and Shi-Tomasi corner detection to separate left and right lung area based on feature points arouncd the lung area,which can find out the adhesion area quickly and accurately.We complete separation through connecting the feature points.Last but not least,because of the sag of lung body caused by some lung diseases,we present the adaptive method based on binary-tree to restore the lung boundaries rapidly.This paper uses parallel adaptive methods to repair the defect area,which can effectively preserve the contour of the normal lung parenchyma sag.We present the lung ROI segmentation algorithm based on Bayesian classification method which apply Gaussian mixture moedel and EM algorithm to compute posteriori probability to get the optimal parameters of image pixels based on Bayesian classification to separate ROI region and background region.The application of the adaptive iterative threshold method as the initial paranmeters of Gaussian mixture model is to avoid increasing the number of iterations and prevent that EM algorithm is sensitive to initial parameters induced by local optimum.In this paper,the segmentation result of ROI is obtained by the Bayesian classification method based on the EM algorithm and the segmentation results of different sub-graphs are analyzed.At the same time,we verify the rationality of GMM through the results of the gray histogram of lung parenchyma image and the iterative result of adaptive iterative threshold method.The optimal distribution of GMM obtained by EM estimation and the fitting situation of the gray histogram of lung parenchymal show that EM iteration algorithm can be well fitted the situation of data missing.