A deep-learning neural network classifier identified patients with clinical heart failure using whole-slide images of tissue with a 99% sensitivity and 94% specificity on the test set, outperforming two expert pathologists by nearly 20%.