Automatic Detection Method Based On Ct Images, Lung Tumor Computer

The mortality rate for lung cancer is higher than that for other kinds of cancersaround the world. At the same time, it appears that the rate has been continuallyincreasing in recent years. Conventional chest radiograms have been used to screenlung cancer and computed tomography is widely used to help detect lung cancerbecause through this, it is possible to diagnosis correctly. However, this process isexhausting for there are so many images that should be interpreted. In addition, theerror of reading could not be avoided. Therefore, computerized automated detectionor diagnosis systems for medical image could be used to help clinicians diagnose,treat, monitor changes, and plan and execute procedures more safely and effectively.For that, many important techniques should be developed, in which there are twofundamental problems, which are segmentation, which localizes the anatomicalstructures in an image, and recognition, which classify the samples into false andpositive.In this thesis, we studied the problem of the detection of nodules in pulmonarycomputed tomography images. The studies are motivated by the potential ofcomputer-aided detection to improve detection accuracy. To make such systempractical, these techniques are required to develop the automated segmentation andrecognition methods.Firstly, adaptive gray-level-thresholding techniques are utilized to segment thelung fields for each pulmonary computed tomography to create a segmented lungvolume. Subsequently, multi-gray-level-thresholding and fuzzy-region-growingtechniques are applied within the segmented lung regions to identify a set of lungnodule candidates. After initial detecting candidates, various objective features onthe nodules were determined by use of outline analysis and image analysis. Weextracted a total of 12 features to train the following classifiers by use of sequentforward search from a total of 29 feature values, which contained the texture ofcandidate region, Fourier description, and global shape. Rule-based classifier andlinear discriminant classifier are used to differentiate between lung nodule candidatesthat correspond to actual nodules and those that correspond to non-nodules. Theautomated method that was applied to 15 cases, which have 70 nodules, could get theaccuracy of 87.1%. For verify the performance of the classifiers, the rule-based...