Classification of pulmonary nodules in thin-section CT images by using multi-scale curvature indexes

摘要: Multi-scale curvature indexes are introduced to characterize the internal intensity structure of pulmonary nodules in thin-section CT images. This approach makes use shape index, curvedness, and density represent locally each voxel constructing three-dimensional (3D) nodule image. Using features extracted from histogram multi-scale density, discriminated between benign malignant cases by linear discriminant classifier. In this study a data set 128 is analyzed investigate which scale provides high classification accuracy nodules. Additionally, evaluated for four different regions: (i) entire 3D nodule; (ii) core region (iii) complement the-core (iv) neighborhood surrounding nodule. The effectiveness computer-aided differential diagnosis demonstrated receiver operating characteristic (ROC) analysis.