Algorithms Research Of Computer-aided Detection Of Solitary Pulmonary Nodules And Medical Signs Recognition

Currently lung cancer has become one of the most frequently-occurring diseases. Early diagnosis and treatment of lung cancer is of great significance for the patients, which can increase 5-year survival rate of patients after surgery to 40% to 70%. The main signs in CT images series of early lung cancer are solitary pulmonary nodules. So the tasks for CT diagnosis of lung cancer are nodules identification in the lung parenchyma, medical signs of nodules observation, and the level of benign and malignant of nodules discrimination. For the patients with benign nodules surgery is not necessary, while for the patients with malignant nodules, early diagnosis and early treatment are very important to save their lives.The main tasks of this thesis are focus on semi-automatic lung nodules detection and medical signs identification. The purpose is to assist the radiologists to identify lung nodules, and to provide a quantitative tool for the radiologists to initially determine the level of benign and malignant of nodules. The contents and the corresponding achievement of this thesis include:1. Based on maximum variance thresholding algorithm, lung images are binarized, lung parenchyma region are detected along scan lines. After the users. select the seed point, lung parenchyma are segmented by region growing algorithm.The experimental results show that, the correct rate of the segmentation is 98% or more.2. By the way of interactively seed point selection, the suspected pulmonary nodules are segmented based on region growing algorithm:then orderly border information are calculated by boundary contour extraction and tracking algorithms; In the meanwhile, some quantitative characteristics of suspected nodules are extracted including geometry, location, intensity and texture characteristics, for the classification of suspected lung nodules which determining the suspected nodules whether be the lung nodules.3. For the medical signs of nodules such as burr, spinous process, calcification, cavitation. empty, and leaf et. the corresponding recognition algorithms are proposed respectively. The experimental results demonstrate that the correct rate of recognition algorithm of burr and spinous process is close to 70%; but the correct rate of recognition algorithm of calcification, cavitation, empty and leaf is above 70%.4. Attribute-Bagging-kNN classification algorithm are first applied in the experimental classification of pulmonary nodule data which is built by the proposed features. The experimental results show that the correct rate of the classification algorithm for pulmonary nodules data is up to 98%, which demonstrate the proposed features of suspected regions has good performance for lung nodule detection.