A Hybrid Preprocessing Method Using Geometry Based Diffusion and Selective Enhancement Filtering for Pulmonary Nodule Detection

The computer aided diagnostic (CAD) system has been developed to assist radiologist for early detection and analysis of lung nodules. For pulmonary nodule detection, image preprocessing is required to remove the anatomical structure of lung parenchyma and to enhance the visibility of pulmonary nodules. In this paper a hybrid preprocessing technique using geometry based diffusion and selective enhancement filtering have been proposed. This technique provides a unified preprocessing framework for solid nodule as well as ground glass opacity (GGO) nodules. Geometry based diffusion is applied to smooth the images by preserving the boundary. In order to improve the sensitivity of pulmonary nodule detection, selective enhancement filter is used to highlight blob like structure. But selective enhancement filter sometimes enhances the structures like blood vessel and airways other than nodule and results in large number of false positive. In first step, geometry based diffusion (GBD) is applied for reduction of false positive and in second step, selective enhancement filtering is used for further reduction of false negative. Geometry based diffusion and selective enhancement filtering has been used as preprocessing step separately but their combined effect was not investigated earlier. This hybrid preprocessing approach is suitable for accurate calculation of voxel based features. The proposed method has been validated on one public database named Lung Image Database Consortium (LIDC) containing 50 nodules (30 solid and 20 GGO nodule) from 30 subjects and one private database containing 40 nodules (25 solid and 15 GGO nodule) from 30 subjects.