Skip to main content
. 2021 Aug 5;2(4):576–585. doi: 10.1093/ehjdh/ztab071

Figure 2.

Figure 2

Overview of our experimental setting: digital biomarkers (HRV and MOR) are engineered from the 12-lead ECG. In addition, a pretrained DNN model is used as a feature generator taking as input the raw 12-lead ECG. Finally, EMR information are added as metadata features. The resulting vector of features (HRV, MOR, DNN, and EMR) is used to train random forest classifier in predicting whether an individual will develop AF within 5 years.