In their study, Crowson et al. evaluate how accurately a neural network algorithm can diagnose pediatric middle ear infections compared to a commercial classifier from Google and human physicians. The researchers trained their deep learning model using 80% of 639 tympanic membrane images collected from pediatric surgery patients diagnosed as normal, effusion (non-purulent), and infection (purulent effusion) by their respective surgeons intraoperatively. The model was then tested on the remaining images and compared to Google’s “AutoML Vision” classifier. Finally, both models were tested on 22 new ear images, and their diagnostic outputs were compared to those of 39 physicians who reviewed the same test images via an online survey. The researchers’ deep learning model achieved a diagnostic accuracy of 80.8% on the initial test images compared to 95.5% on the survey images, indicating some inherent variance, while Google’s model achieved around 86% accuracy on both sets. On the survey images, pediatricians and general physicians had 60% accuracy overall, while otolaryngologists achieved about 80% accuracy. Both deep learning models struggled to distinguish between non-purulent effusions and infections in the ear; however, they outperformed most physicians. The authors highlight that machine learning tools can significantly reduce diagnostic mistakes and improve the management of middle ear infections for patients.