Ha R, et al. - Researchers evaluated if convolutional neural networks could predict which patients with pure atypical ductal hyperplasia (ADH) can be monitored safely vs undergoing surgery. For the convolutional neural network algorithm, a total of 298 unique images from 149 patients were used. Data reported that the AUC value was 0.86 for the test set. Investigators found that aggregate sensitivity was 84.6% and specificity was 88.2% and diagnostic accuracy was 86.7%. Using mammographic images, it is possible for convolutional neural networks to discriminate pure atypical ductal hyperplasia from ductal carcinoma in situ. Further improvement in the prediction model would likely result from a larger dataset.