Fig. 3: Structure and performance of ECG-WMA-Net for detecting WMAs from the ECG. | npj Digital Medicine

Fig. 3: Structure and performance of ECG-WMA-Net for detecting WMAs from the ECG.

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

Fig. 3

a Structure of ECG-WMA-Net – Input to the model was the first 5 s of the eight unique leads from the ECG recording. At 500 Hz, this resulted in an input matrix of (1, 2500, 8). The parameter searching tool optimally found six sets of Conv, BatchNorm, Max Pool, and Dropout prior to Flatten, and Dense layers. b Receiver operating characteristics of ECG analysis by ECG-WMA-Net (yellow), quantitative ECG analysis (blue), and qualitative ECG analysis (green) for detection of WMA on echocardiography. The ‘x’ for each curve corresponds to the Youden Index optimal cut point. c Receiver operating characteristics of ECG analysis by ECG-WMA-Net in the internal test cohort (yellow) and external test cohort (blue).

Back to article page