Research On Pulmonary Parenchyma Segmentation Method Based On CT Image

Recently,computer tomography technique(Computed Tomography,CT)medical imaging technology has been developing continuously.Its density resolution is high,easy to operate,non-invasive,so it has the extensive application in human body important organ examination and the medical diagnosis.The World Health Organization(world health organization,WHO)has shown that lung cancer has been the most common form of cancer that poses a serious threat to human health.The main manifestations of pulmonary abnormalities are calcification,lumps,cavities and bronchiectasis on CT images of multiple subjects,and lung segmentation is a key step in any clinical decision-making aimed at improving the early diagnosis and treatment of lung diseases.Although the research of lung segmentation at home and abroad have been very mature,there are still many problems such as accuracy and robustness of segmentation.In many kinds of medical images,CT can show the morphological changes of lung tissue and interstitial structure more clearly.Therefore,the pulmonary parenchyma segmentation and extraction of lung CT images can help doctors to locate and analyze the lesion location better,to diagnose lung diseases in order to take more targeted treatment programme.This paper analyzed the problems of the existing lung parenchyma segmentation algorithms,an improved median filter preprocessing method is proposed,which provides a good basis for the complete segmentation of lung parenchyma.Then two different lung parenchyma segmentation methods(including pretreatment,rough segmentation,fine segmentation,left and right lobe separation,edge repair and other steps)are proposed.Finally,through the comparison with the existing mature algorithm,and make a corresponding evaluation.The main research contents of this paper are as follows:(a)An improved median filtering method based on the main noise interference in CT image is proposed.In this paper.Firstly,the impulse noise of the image is judged.According to the gray range of the foreground and noise in the image and the relation between them(including the criterions of detecting the noise point),the suspected impulse noise(that is,the salt and pepper noise)is searched,and these suspected pepper and salt noise points are removed.Then the non-pepper and salt noise points are reordered and formed a new set.After multiple searches,sort combinations.Finally,according to the estimated global noise adaptive selection required filter window size,then all the identified noise points can be filtered to complete the improved median filtering algorithm.(b)Proposed an automatic segmentation algorithm for lung parenchyma repair based on optimal threshold and local rolling ball method.Firstly,the obtained DICOM CT images are converted into gray scale,then the lung parenchyma is filled with chest contour by using the improved median filtering and image enhancement preprocessing,use the optimized threshold binarized the image,then to fill the chest contour.The initial mask of lung parenchyma was obtained by removing the small area in the background by mathematical morphology operation.Then it is marked as two small areas that satisfy different characteristics by the connected domain,and the left and right lung lobes are separated by row and column scanning to complete the initial segmentation of the lung parenchyma.Finally,the edge of the lung parenchyma is repaired by the "local" rolling ball method.The final lung parenchyma segmentation results can be obtained by point multiplication with the pre-processed image.(c)An improved OTSU(maximum inter-class variance method)combined with morphological-assisted operation is proposed to extract lung parenchyma.Firstly,the original CT image is pre-processed and binarized by median filtering and image enhancement,the chest mask is obtained,and the initial mask of the lung is obtained by the improved OTSU algorithm.Through the method for determining the upper and lower end points of the left and right lung lobes to find a straight line which is used as a boundary to realize the separation of the left and right lung lobes.Then,the morphology is used to smooth it,remove the sharp prickles.Finally,they are combined and filled with lung parenchyma mask,so that the final lung parenchyma segmentation image can be realized by adopting a multiplication operation on the pre-processed original image.The simulation results show that under the condition of keeping the complexity of the existing image segmentation algorithms almost unchanged,the segmentation accuracy is improved and the original defects are overcome.(d)Analysis of the experimental results.The DICOM format CT sequence images in the data set of the Lung Image Database Alliance(Lung Image Database Consortium,LIDC),sponsored by the National Cancer Institute(National Cancer Institute)were used in the experiment.The results of two kinds of lung parenchyma segmentation were observed subjectively by naked eyes,objective data analysis:segmentation accuracy,recall rate,Dice coincidence rate and other image segmentation evaluation indexes,the two algorithms and the existing segmentation methods were compared.According to statistics,the segmentation accuracy of the optimized threshold combined with the local rolling ball method and the improved OTSU combined morphology is 98.53%,97.56%,respectively;and the recall rate is 97.65%and 99.29%,respectively;Dice coincidence rate is 98.17%and 98.42%,respectively.It is proved that the optimization and improvement of the existing segmentation algorithms in this paper,and fully reflect the effective and complete extraction of the lung parenchyma in lung CT image is proved to be the application value in the medical image processing.