Research Of Lung Parenchyma Segmentation Based On CT Images

Lung cancer is one of the common malignant tumors,and with the highest morbidity and mortality in our country.The key to reducing lung cancer mortality is early detection,early diagnosis and early treatment.Nodules are one of the most important signs of lung disease,so the detection of pulmonary nodules is very important.As one of the important methods for detecting and diagnosing pulmonary diseases,CT scanning technology has high spatial resolution and has been used for the clinical diagnosis of pulmonary diseases.However,the number of images obtained by CT scans is huge,and the organizational structure is complex.There are many types of lung diseases,so clinical testing results are not ideal.In order to solve these problems,domestic and foreign scholars have proposed a combination of multiple algorithms,but the results still can not meet the clinical needs.Therefore,in this paper,the CT images of the lungs are used as the data source to conduct in-depth research on how to improve the accuracy of image segmentation.Firstly,this paper proposes an automatic lung image segmentation method based on random walk.In this method,the image is preprocessed.For the features of low contrast and weak edges of CT images of lungs,the image brightness is enhanced by image enhancement,threshold segmentation,etc.,and the non-pulmonary regions such as the scan bed are removed to obtain lungs.Substantial binary image.Then the image is segmented based on an improved random walk algorithm.Compared with the traditional algorithm,this algorithm improves the edge weight in the process of converting the image into a graph,which makes the walking probability between two adjacent pixels.Affected by the grayscale value of the pixel,it is also affected by the distance.In addition,this paper select the seed points by finding the center of gravity of the connected regions automatically,and finally realizes the automatic segmentation of the lung parenchyma.And then use the rolling ball method to repair the lung parenchyma,the obtained lung parenchymal images were compared with the gold standard.It was found that the segmentation accuracy of this segmentation method was as high as 99%for normal lung images,but the segmentation accuracy of lung images with narrower and larger depressions in the right and left lung mediastinum was lower,and not only complete lung parenchyma could not be obtained,but also lost the key information needed for diagnosis.For this algorithm,the complete lung parenchyma could not be obtained.For the above defects,this paper were respectively repaired using corner detection,two-dimensional convex hull and vector method.By comparing the repair results,It was found that the accuracy of the repaired results using the improved vector method was 0.9%better than that of the simple rolling ball method,and it retained the necessary information for diagnosis well.Since the random walk algorithm segment images based on a single pixel point,it is relatively slow,especially for the images with more complex surrounding tissue,the result is not ideal,so the fourth chapter based on the superpixel image Refined segmentation,this method can accurately segment the lung parenchyma from the surrounding tissues.This paper first uses Turbopixel method to form superpixels,and then use FCM to classify pixels,which improves the segmentation efficiency,retains more edge information,and ensures high clinical application value.