Automatic segmentation of the lung fields in portable chest radiographs based on Bézier interpolation of salient control points

Abstract

Portable chest radiographs are performed on the most critically-ill patients, who need highly skilled patient care and the best diagnostic tools. However, such radiographs are usually of low quality, mainly due to misaligned body positioning during their acquisition, and subject to a higher misinterpretation rate than the ones obtained with a non-portable X-ray device. This paper presents a pioneering computational approach that copes with the segmentation of the lung fields in portable chest radiographs. The proposed methodology involves detection of salient points on the anatomic structures around the lung fields by subsequent application of simple intensity and edge feature extraction techniques. The salient points detected are interpolated using Bezier curves intuitively approximating the boundaries of the lung fields. Unlike current methodologies, the proposed one does not exclude the overlapping region of the heart from the lung fields, where abnormalities can also be present. The results illustrate the robustness of the proposed methodology on a set of real portable radiographs of patients with bacterial pulmonary infections. A qualitative comparison with a state of the art approach based on graph cuts validates its effectiveness.