Pulmonary nodule detection using cartwheel projection analysis

摘要

In this paper, we propose a novel method, called cartwheel projection analysis (CWPA), to automatically detect lung nodules in high-resolution multi-slice CT images. First, a nodule candidate generation algorithm is applied to examine the whole lung volume and record only structures that are potentially nodules. This step quickly excludes small and large sized vessels from further analysis, so that computation can be saved dramatically. Then, cartwheel projection analysis is applied to examine the shape characteristics of the structures corresponding to the nodule candidates. CWPA is based on 1-dimensional curves obtained from a series of 2D cutting planes centered at the structure of interest in the volume. Finally, a set of criteria is applied for identifying the existence of nodules based on shape analysis of the 1-D shape curves.