Local pulmonary structure classification for computer-aided nodule detection

Abstract

ABSTRACT We propose a new method of classifying the local structure types, such as nodules, vessels, and junctions, in thoracic CTscans. This classiÞcation is important in the context of computer aided detection (CAD) of lung nodules. The proposedmethod can be used as a post-process component of any lung CAD system. In such a scenario, the classiÞcation resultsprovideaneffectivemeansofremovingfalsepositivescausedbyvesselsandjunctionsthusimprovingoverallperformance.Asmainadvantage, theproposedsolutiontransformsthecomplexproblemofclassifyingvarious3Dtopologicalstructuresinto much simpler 2D data clustering problem, to which more generic and sexible solutions are available in literature, andwhich is better suited for visualization. Given a nodule candidate, Þrst, our solution robustly Þts an anisotropic Gaussianto the data. The resulting Gaussian center and spread parameters are used to afÞne-normalize the data domain so asto warp the Þtted anisotropic ellipsoid into a Þxed-size isotropic sphere. We propose an automatic method to extract a3D spherical manifold, containing the appropriate bounding surface of the target structure. Scale selection is performedby a data driven entropy minimization approach. The manifold is analyzed for high intensity clusters, corresponding toprotrudingstructures. TechniquesinvolveEMclusteringwithautomaticmodenumberestimation,directionalstatistics,andhierarchical clustering with a modiÞed Bhattacharyya distance. The estimated number of high intensity clusters explicitlydetermines the type of pulmonary structures: nodule (0), attached nodule (1), vessel (2), junction (>3). We show accurateclassiÞcation results for selected examples in thoracic CT scans. This local procedure is more sexible and efÞcient thancurrent state of the art and will help to improve the accuracy of general lung CAD systems.Keywords: Computer-Aided Diagnosis, CT, Pulmonary Nodules, Statistical Clustering, Directional Statistics