Artificial Intelligence-Enabled Electrocardiogram Predicted Left Ventricle Diameter as an Independent Risk Factor of Long-Term Cardiovascular Outcome in Patients With Normal Ejection Fraction

Figures

• 

  FIGURE 1. Development, tuning, internal validation, and external validation sets generation and the ECG labeling of echocardiogram. Schematic of the dataset creation and analysis strategy, which was devised to assure a robust and reliable dataset for training, validating, and testing of the network. Once a patient’s data were placed in one of the datasets, that individual’s data were used only in that set, avoiding “cross-contamination” among the training, validation, and test datasets. The details of the flowchart and how each of the datasets was used are described in “Materials and Methods” section.

• 

  FIGURE 2. Scatter plots of predicted value via ECG voltage–time traces only compared with the actual ejection fraction (EF) and left ventricle (end-diastole/end-systole) (LV-D/LV-S). The x-axis indicates the actual value and the y-axis presents the ECG-predictions. Red points represent the highest density, followed by yellow, green light blue, and dark blue. We presented the mean difference (Diff), Pearson’s correlation coefficients (r), and mean absolute errors (MAEs) to demonstrate the accuracy of a deep learning model (DLM). The black lines with 95% conference intervals (CIs) are fitted via simple linear regression.

• 

  FIGURE 3. Receiver operating characteristic (ROC) curve analysis for mild to severe left ventricle abnormality from deep learning model based ECG voltage–time traces. The ROC curve (x-axis = specificity and y-axis = sensitivity) and area under ROC curve (AUC) were calculated using the internal validation set (A) and external validation set (B). The operating point was selected based on the maximum of Yunden’s index in tuning set, which was used for calculating the corresponding sensitivities and specificities in two validation sets.

• 

  FIGURE 4. Stratified analysis for model performance for predicting electrocardiogram-based ejection fraction (ECG-EF), left ventricular end-diastolic diameter (LV-D), and left ventricular end-systolic diameter (LV-S). The analyses were stratified by the disease histories or data source. The AUC and 95% CIs were presented based on LV EFs and diameters. ER, Emergency room; IPD, inpatient department; OPD, outpatient department; DM, diabetes mellitus; HTN, hypertension; HLP, hyperlipidemia; CKD, chronic kidney disease; AMI, acute myocardial infarction; STK, stroke; CAD, coronary artery disease; HF, heart failure; Afib, atrial fibrillation; COPD, chronic obstructive pulmonary disease (COPD).

• 

  FIGURE 5. Scatter plots of predicted and actual EF correlated with LV-D. The x-axis indicates the actual EF and the y-axis presents the ECG-EF. Green to red points represent the small and large predicted and actual LV-D, respectively. The percentages were the proportion of people with an ECHO/ECG LV-D > 56 mm in each ECHO and ECG EF group.

• 

  FIGURE 6. Relationship between most important ECG features and ECG estimated left ventricle (end-diastole) (ECG-LV-D). The related importance is based on the information gain of eXtreme gradient boosting (XGB) model, and the R-square (R-sq) is the coefficient of determination to use selected ECG features for predicting ECG-LV-D. The AI-ECG predictions were classified as ECG-normal (ECG-LV-D ≤ 56) and ECG-abnormal (ECG-LV-D > 56). The analyses are conducted both in internal and external validation sets (***p < 0.001).

• 

  FIGURE 7. Long-term incidence of developing a new-onset left ventricular dysfunction (LVD, EF ≤ 35%) in patients with an initially normal EF (EF > 50%) stratified by ECG-EF, LV-D, and ECG-LV-D. The C-index is calculated based on the continuous value combined with sex and age. The analyses are conducted both in internal and external validation sets. The table shows the at-risk population and cumulative risk for the given time intervals in each risk stratification.

• 

  FIGURE 8. Risk matrixes of different predicted ejection fraction (ECG-EF) and left ventricle (end-diastole) (ECG-LV-D) groups on adverse events in patients with an initially normal EF (EF > 50%). The hazard ratios (HRs) are based on the Cox proportional hazard model adjusted by gender and age. The color gradient represents the risk of corresponding group, and the non-significant results are defined as white.