A nonparametric-based rib suppression method for chest radiographs

Abstract

This paper presents an automated and comprehensive system for eliminating rib shadows in chest radiographs, which integrates lung field identification, rib segmentation, rib intensity estimation, and suppression. We designed a region of interest (ROI)-based method to estimate a suitable initial lung boundary for active shape model (ASM) deformation by determining the translation and scaling parameters from the lung ROI. By considering the anatomical structure of the rib cage, we developed a locale sampling scheme to achieve nonparametric rib modeling. This scheme integrates knowledge-based generalized Hough transform (GHT) for accurate rib segmentation. We subsequently estimated rib intensity using the real-coded genetic algorithm (RCGA). Experimental results indicate that the relative conspicuity of the nodules increased after rib suppression, compared to the original image. Additionally, the proposed system uses only one standard chest radiograph, and the dual-energy subtraction technique is not required. Thus, this system is suitable for radiologists and computer-aided diagnosis (CAD) schemes for detecting lung nodules in chest radiographs.