Fig. 2: Diagram and performance of the deep learning-based algorithm for identifying patients with active AF or paroxysmal AF at the time of TTE. | npj Digital Medicine

Fig. 2: Diagram and performance of the deep learning-based algorithm for identifying patients with active AF or paroxysmal AF at the time of TTE.

From: Deep learning evaluation of echocardiograms to identify occult atrial fibrillation

Fig. 2

Using two serial convolutional neural networks, TTEs were classified as being in active AF or in sinus with paroxysmal AF within 90 days, both scenarios where anticoagulation or rhythm control would be potentially indicated. TTEs were first stratified as being in AF or sinus rhythm. The model first determined AF versus sinus rhythm with an AUROC of 0.96 (95% CI 0.95–0.96). From the subset in the sinus, the model then further predicted which TTEs had concurrent paroxysmal AF, defined as having AF on ECG within 90 days before or after, with an AUROC of 0.74 (0.71–0.77).

Back to article page