Curvature-based automated detection of pulmonary nodules in CT images with surface feature-based false positive reduction

We propose a shape-based automated detection method for the detection of pulmonary nodules with surface feature-based false positive (FP) reduction. In the proposed system, shape index-based thresholding and region-growing are first performed to segment nodule candidates. According to the sizes of the nodules, multi-scale integration based on the eigenvalue of the Hessian matrix is employed to obtain the appropriate shape index. In the FP reduction step, FPs at bifurcations of vessels are removed using the extracted surfaces of the vessels and nodules. We evaluated the proposed system using 16 chest CT scans that included nodules identified by two experienced radiologists. The results of this evaluation showed that the proposed FP reduction scheme is effective for reducing FPs at vessel bifurcations. (author)