Automated multilabel diagnosis on electrocardiographic images and signals

Figures

• 

  Fig. 1. Study outline for waveform preprocessing and image transformation for modeling. ECG electrocardiogram.

• 

  Fig. 2. Area under the receiver operator characteristic (AUROC) curves for image and signal-based models. AUROC on held-out test set (A, image model; B, signal model) and cardiologist validated test set (C, image model; D, signal model) are included in the 4 image panels. 1dAVB 1st degree AV block, AF atrial fibrillation, LBBB left bundle branch block, RBBB right bundle branch block, SB sinus bradycardia, ST sinus tachycardia.

• 

  Fig. 3. Area under the receiver operator characteristic (AUROC) curves on the external validation set (PTB-XL). A Image based model, and B Signal based model. 1dAVB 1st degree AV block, AF atrial fibrillation, LBBB left bundle branch block, RBBB right bundle branch block, SB sinus bradycardia, ST sinus tachycardia.

• 

  Fig. 4. Confusion matrices for image model predictions. A Held-out test set, B cardiologist-validated test set, and C PTB-XL. 1dAVB 1st degree AV block, AF atrial fibrillation, LBBB left bundle branch block, RBBB right bundle branch block, SB sinus bradycardia, ST sinus tachycardia.

• 

  Fig. 5. Gradient-weighted Class Activation Maps for Image Predictions. Average of the Gradient weighted Class Activation Maps (Grad-CAMs) for the 25 most confident predictions of right bundle branch block (RBBB) and left bundle branch block (LBBB) on standard and alternate format images. A RBBB in standard format, B RBBB in alternate format, C LBBB in standard format, and D LBBB in alternate format.