Lung cancer is one of the deadliest kinds of cancer. Pulmonary nodules are one of the symptoms of lung cancer that can appear as a separated mass or as a mass attached to chest wall. Nodule is an additional mass that is produced inside the lung. Lung cancer is not detected in early stages in computed tomography (CT) scan images. Pulmonary nodules appear when lung cancer is in the more severe stages, because of low contrast, small size and position of the pulmonary nodules in CT scan images. When a nodule is small, its detection by eye can be difficult. For this reason, computerized image processing algorithms can help the radiologist in the detection, localization and quality assessment of nodules by highlighting the suspicious areas. Computer aided detection (CAD) systems are used as an intelligent tool that express a second opinion for the radiologist. CAD systems can represent the suspicious areas in CT scan images to the radiologist to be more carefully studied. The basic idea is not to design a software to make the final decision, but to make a software to help the radiologist in his diagnosis. Accordingly, in this thesis, a new algorithm is developed to detect the suspicious nodule candidates in CT scan images. Firstly, lung lobe regions are segmented from the original images to increase the processing speed and accuracy. In the next stage, template matching method is used to detect the suspicious nodule candidates. These suspicious nodule candidates include nodules and some vessels that are similar to pulmonary nodule in single frames. Then we segment the suspicious nodule candidates by localized active contours. Using a single frame leads to many false positive errors. So, our algorithm will use consecutive slices to enhance the detection process. Nodules and blood vessels have spherical and cylindrical shape, respectively. There is a small difference between large diameter and small diameter in pulmonary nodules, so three-dimensional and two-dimensional features are extracted from areas of suspected pulmonary nodules that represent the physical dimention the suspicious nodule candidates. Finally, for left; TEXT-INDENT: 36pt; MARGIN: 0cm 0cm 0pt; unicode-bidi: embed; DIRECTION: ltr" align=left Keywords: Detection of Pulmonary Nodules, Processing of CT Scan Images, Active Contour, Maximum Correlation, Feature Extraction, Classification
