ERS transform for the automated detection of bronchial abnormalities on CT of the lungs

The identification of bronchi on Computed Tomography (CT) images of the lungs provides valuable clinical information in patients with suspected airways diseases including bronchiectasis, emphysema, or constrictive obliterative bronchiolitis. The automated recognition of the airways is, therefore, an important part of a diagnosis aid system for resolving potential ambiguities associated with intensity-based feature extractors. On CT images, near-perpendicular cross sections of bronchi normally appear as elliptical rings and this paper presents a novel technique for their recognition. The proposed method, the edge-radius-symmetry (ERS) transform, is based on the analysis of the distribution of edges in local polar coordinates. Pixels are ranked according to local edge (E) strength, radial (R), uniformity and local symmetry (S). A discrete implementation of the technique is provided which reduces the computational cost of the ERS transform by using a geometric approximation of the intensity patterns. The identification of the adjacent pulmonary vessels with template matching then allows for the automated measurement of bronchial dilatation and bronchial wall thickening. Computationally, the method compares favorably with other methods such as the Hough transform. Noise-sensitivity of the technique was evaluated on a set of synthetic images and 9 patients under investigation for suspected airways disease. Agreement for the automated scoring of the presence and severity of bronchial abnormalities was demonstrated to be comparable to that of an experienced radiologist (kappa statistics κ>0.5).