Classifier design for computer-aided diagnosis: Effects of finite sample size on the mean performance of classical and neural network classifiers