Abstract: Deep neural networks often require large-scale, accurately labeled datasets to perform well, but in practice the labels are frequently corrupted by noise in medical imaging, especially ...