关键词: 肺实质分割, ROI提取, 决策规则, Mahalanobis距离矢量, 混合分类

Abstract: The approach to weight segmentation with optimum threshold values was used to get the pulmonary parenchyma, and the regions of interests (ROI) were obtained by C-means clustering algorithm and classified by hybrid method, during which two three-dimensional characteristics of pulmonary nodules and two corresponding rules were defined for regular initial classification. Then, a nonlinear classifier was constructed by improving the Mahalanobis distance vector to classify ROI again, so as to lower the false positive. After the hybrid classification, the pulmonary nodules were differentiated efficiently from such interfering information as bloods. Experiment results indicated that the algorithm is highly sensitive to detecting pulmonary nodules.