Skip to main content
. 2022 Sep 9;5:138. doi: 10.1038/s41746-022-00674-x

Fig. 5. ON/OFF accuracy of deep-learning models and human raters.

Fig. 5

Only the ON/OFF accuracies for the hyperparameter combinations that yielded the best CNN and GAN discriminator performance are plotted. Models were tested with ten Study 1 development set subjects and nine Study 2 subjects. The ON/OFF accuracies of the in-person clinician rater are also plotted, alongside the performance of two human raters who assessed 25 Study 1 patients over video.