The Study Of Pulmonary Nodule Detection Method Based On CT Images

Lung cancer is currently the highest mortality in many countries visceral tumors,each year about 1.3 million people in the world wide die of lung cancer,and its incidence is rising year by year.It has been ranked firstly in cancer mortality in many cities in the world.In computer-aided detection and diagnosis of cancer,lung CT images have improve a very important role,it can improve the diagnostic efficiency and quality,and the radiologist reading from a large slice freed.Lung CAD computer-aided diagnosis system can automatically identify lung nodules in CT images,it has play a very important role the treatment of lung cancer,and it has become a hot spot of global research.Based on the CT-depth study of lung cancer CAD computer-aided diagnosis system in the lung parenchyma extraction method,the candidate nodule detection method and method for removing false positives.Traditional region growing need for manual select seeds,select seeds for larger image segmentation results.In this paper,three-dimensional region growing,small regional connectivity removal methods,morphological techniques and bump bypass method,proposed a simple and effective segmented way of parenchyma of lung from a full set of CT images in the sequence.Search first rectangular area from the center of the first full set of CT images(one third of the area of the entire CT)inside the trachea similar circular area,select one seed point to the next point of the three-dimensional region growing step by step,to get three-dimensional lung area;then traverse the image a three-dimensional region growing layer by layer after removing each less than a certain area of the small regional connectivity to eliminate the trachea and bronchus;followed by morphological opening operation to eliminate holes in the lung area;final projection repair lung bypass border point method,the final phase CT images obtained with the original lung parenchyma.Mean shift algorithm is an iterative search to a local density maximum point segmentation method,the method is simple and good segmentation,clustering,image smoothing,image segmentation,and so has a very wide range of applications.In this paper,the method is applied to the candidate nodule detection,and for its noise,local details are very sensitive,prone to over-segmentation phenomenon made two improvements:(1)the traditional mean shift image segmentation algorithm based on membership in the window the size of the rule set to adaptive,so that less band width in high-density areas in the low-density area larger bandwidth;(2)pulmonary nodulesegmentation using mean shift algorithm initially,for each candidate region,with its centroid as the center,use adaptive threshold method in an area twice the matrix area in search of candidate nodule boundaries again to get a more accurate candidate nodule area boundary,for subsequent nodule feature extraction step to lay the foundation.In this paper,SVM support training machine candidate nodules classification,removing false-positive nodules.And nodules feature extraction stage addition of two new features: radial gradient and the candidate nodule density ratio.Not only that the radial gradient of the geometric relationship of a point within and surrounding the nodule candidate points,but also reflects the relationship between the gray level of the region,the application can distinguish between the characteristics of the circular-like structure(pulmonary nodules)and non-circular configuration(e.g.blood vessels).Candidate nodule density ratio refers to the nodule candidate to the center of the centroid of the area of a certain rectangular area,a high density of pixels than the number of the dots of the dot density of the number of pixels except the nodule candidate region.Experiments show that these two new features of the nodules and pulmonary nodules with better distinction.