Convolutional neural networks (CNNs) trained with 20,000 labeled images can accurately classify chest x-rays as normal or abnormal, according to new findings published in Radiology. Training the CNN with an additional 180,000 images, the authors noted, only yielded “marginal” benefits.