Figure 4
Performance of the deep-learning models for classification of atrial fibrillation based on 30-s segment data and whole-length recording of subjects. AF atrial fibrillation, NSR normal sinus rhythm, ROC receiver operating characteristic. Receiver operating characteristic (ROC) curves based on data of 30-s segments (Left) and voting results of whole-length recording of subjects (Right) in the datasets: “AF vs NSR” (Fig. 2A,B), “AF vs Others” (Fig. 2C,D), “AF vs Non-AF” (Fig. 2E,F). By voting algorithm, subject who had more than 50% of model-determined AF segments was classified as AF case.
