Differentiation between nodules and end-on vessels using a convolution neural network architecture.
In recent years, many computer-aided diagnosis schemes have been proposed to assist radiologists in detecting lung nodules. The research efforts have been aimed at increasing the sensitivity while decreasing the false-positive detections on digital chest radiographs. Among the problems of reducing the number of false positives, the differentiation between nodules and end-on vessels is one of the most challenging tasks performed by computer. Most investigators have used a conventional two-stage pattern recognition approach, ie, feature extraction followed by feature classification. The performance of this approach depends totally on good feature definition in the feature extraction stage. Unfortunately, suitable feature definition and corresponding extraction implementation algorithms proved to be very difficult to define and specify. A convolution neural network (CNN) architecture, trained by direct connection to the raw image is proposed to tackle the problem. The CNN, which uses locally responsive activation function, is directly and locally connected to the raw image. The performance of the CNN is evaluated in comparison to an expert radiologist. We used the receiver operating characteristics (ROC) method with area under the curve (Az) as the performance index to evaluate all the simulation results. The CNN showed superior performance (Az = 0.99) to the radiologist's (Az = 0.83). The CNN approach can potentially be applied to other applications, such as the differentiation of film defects and microcalcifications in mammography, in which the image features are difficult to define or not known a priori.