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