Multiple high-regional-incidence cardiac disease diagnosis with deep learning and its potential to elevate cardiologist performance

Figures

• 

  Figure 1. ROC curves of prediction sensitivity of our deep learning model for the included rhythm types on the test dataset ROC = receiver operating characteristic.

• 

  Figure 2. Confusion matrix for the predictions of our deep learning model versus the clinical diagnosis result on the test dataset (A–O) = normal sinus rhythm, B = sinus tachycardia, C = sinus bradycardia, D = sinus rhythm & T wave changes, E = sinus arrhythmia, F = sinus rhythm & ST-T segment changes, G = sinus rhythm & complete right bundle branch block, H = sinus rhythm & premature atrial contraction, I = atrial fibrillation, J = sinus rhythm & left ventricular high voltage, K = sinus rhythm & premature ventricular contraction, L = sinus rhythm & left ventricular hypertrophy with ST-T segment changes, M = sinus tachycardia & ST-T segment changes, N = sinus bradycardia & T wave changes, O = atrial fibrillation & ST-T segment changes, p = paced rhythm.

• 

  Figure 3. Confusion matrix for the predictions of our deep learning model versus the board-certified final result on the re-annotated dataset

• 

  Figure 4. Confusion matrix for the predictions of the reference group versus the board-certified final result on the re-annotated dataset

• 

  Figure 5. Confusion matrix for the predictions of the non-reference group versus the board-certified final result

• 

  Figure 6. ROC curves of prediction sensitivity of our deep learning model and cardiologists for the included rhythm types on the re-annotated dataset

• 

  Our deep learning model architecture. The neural network architecture used for ECG classification