Fig. 1: Process for development and external testing of ECG-WMA-net using ECG and echocardiogram pairs from Stanford and Emory University cohorts. | npj Digital Medicine

Fig. 1: Process for development and external testing of ECG-WMA-net using ECG and echocardiogram pairs from Stanford and Emory University cohorts.

From: Identification of cardiac wall motion abnormalities in diverse populations by deep learning of the electrocardiogram

Fig. 1

Consecutive ECG-echocardiogram pairs were curated from the clinical ECG and echocardiogram databases at Stanford (left) and Emory (right) Universities. Pairs were excluded if the two studies were >60 days apart, if the ECG reported ventricular pacing, or if ECG data were missing. Only the most recent valid pair was used such that all patients in the database were unique. At Stanford, semi-structured echocardiogram reports and ECG reports were parsed by NLP for WMA labels and ECG labels, respectively. Raw ECG data were used with WMA labels to train ECG-WMA-Net for comparison against the standard of care. Consecutive ECG-echocardiogram pairs from Emory were then used to externally test the model.

Back to article page