Computer-aided diagnosis of pulmonary nodules in chest radiographs: a wavelet-based snake approach

Abstract

A wavelet-based snake model has been developed for distinction between nodules and false positives reported by our computer-aided diagnosis (CAD) scheme for detection of pulmonary nodules in digital chest radiographs. In our method, the boundary of a nodule is first approximated by multi-scale edges, which are then used to guide the snake to estimate the true boundary of the nodule. The deformation of the snake is performed by a maximum likelihood (ML) estimate using a gradient descent algorithm based on the fast wavelet transform. The degree of overlap between the fitted snake and the multi-scale edges was used as a measure for distinction between nodules and false positives. A set of regions of interest (ROIs) consisting of 84 nodules and 694 false positives were used for evaluation of our method by means of the receiver operating characteristic (ROC) analysis. The wavelet snake alone yielded an area under the ROC curve (Az) of 0.75 in discrimination performance. When combined with other 10 morphological features, the performance was increased to Az=0.80, whereas the Az value obtained with these morphological features alone was 0.75.