Figure 2.

Performance of physicians compared to deep learning classifier

We trained two classifiers, one on focused images and the other on panoramic images. We compared the performance of the classifiers to that of pediatricians and medical geneticists. In the boxplots, each point represents the accuracy difference between the classifier and the human performance for a single survey, with the ranges for each group of respondents shown by the lines extending from each boxplot. The red line indicates the baseline accuracy for the classifier.