Fig. 3: AF prediction performance using deep learning compared to other risk prediction methods.
From: Deep learning evaluation of echocardiograms to identify occult atrial fibrillation
Performance of a deep learning model for prediction of concurrent paroxysmal AF from PLAX TTE videos in sinus rhythm compared to models using CHARGE-AF clinical risk factors (age, race, height, weight, hypertension, smoking history, diabetes, heart failure, and myocardial infarction history), PLAX measurements (age, sex, LA diameter, LV end-diastolic and systolic diameters, septal diameter, LV posterior wall diameter), left atrial size, and CHA2DS2VASc score. Model positive predictive value (PPV) and number needed to screen (NNS) with 95% confidence intervals at different sensitivity thresholds are presented.
