肺部CT图像ROIs的分割及肺结节检测-计算机科学与技术专业论文.docxVIP

哈尔滨工业大学工学硕士学位论文 哈尔滨工业大学工学硕士学位论文 - - II - Abstract Lung cancer is still one of the cancers threating human life and health, so early detection of lung cancer will help improve the survival rate of patients with lung cancer. Relying solely on the subjective diagnosis of doctors on large number of medical CT images, the fatigue or other factors often cause missed diagnosis. Automatic detection programs of lung nodules can help doctors improve efficiency and reduce the intensity of work. Accurate segmentation of lung nodules is demanded to accurately detect nodules, reduce false positive rate of identification and classification and evaluate the clinical treatment effect. Based on application background of Computer-aided diagnosis and treatment evaluation system, this paper studies the segmentation of the ROIs in the lung CT images and detection of suspected nodules. First, the DICOM files of CT images are covered to facilitate computer processing. For the preprocessed images, region growing algorithm based on Otsu is applied to obtain lung parenchyma, and the SUSAN (Smallest UnivalueSegment Assimilating Nucleus) operator is used to compensate lung parenchyma. Then according to the image intensity distribution, apply Gaussian mixture model (GMM) and Bayesian classifier method to realize binary segmentation of lung parenchyma images to obtain the initial ROIs. Also, dividing the image into the same size patch, employ iteration algorithm with adaptive threshold value to initialize the parameters of GMM and apply EM algorithm to estimate GMM optimal parameters. Second, due to the gray difference between background and target objects, we proposed edge expansion algorithm based on gradient information for accurate segmentation of ROIs according to the initial shape of ROIs. This paper exploits Sobel operator combined with the regional boundary gradient to detect and locate edge until obtain the approximate real boundary of ROI. In this paper, the proposed method is tested in