Personalized artificial intelligence based left ventricular ejection fraction and systolic dysfunction assessment.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Left-ventricular ejection fraction (LVEF) is crucial for assessing cardiac function.
• Transthoracic echocardiography (TTE) is the standard but resource-intensive method for LVEF assessment.
• Accessible alternatives for LVEF estimation are needed.

Purpose of the Study:

• To evaluate the electrocardiogram (ECG) as a tool for estimating LVEF.
• To develop and validate AI models for LVEF estimation using ECG data.
• To assess the utility of ECG-based models for screening left ventricular systolic dysfunction.

Main Methods:

• Development of convolutional and probabilistic neural network models using ECG data from 191,941 patients.
• Comparison of ECG-only models versus hybrid models incorporating structured features.
• Validation of model performance using mean-absolute-error (MAE) and root-mean-square-error (RMSE).
• Evaluation of model performance in identifying LV systolic dysfunction (LVEF ≤ 40%) using AUC, sensitivity, and NPV.

Main Results:

• ECG-only models achieved MAE of 7.71% and RMSE of 10.36%.
• Hybrid models achieved MAE of 7.84% and RMSE of 10.52%.
• Personalized models significantly improved accuracy (MAE as low as 5.98%).
• LV systolic dysfunction screening achieved AUC of 0.88, sensitivity of 0.92, and NPV of 0.98.

Conclusions:

• ECG-based models can accurately estimate LVEF and quantify uncertainty.
• These models offer a promising, accessible method for LVEF assessment.
• ECG analysis demonstrates high performance in screening for LV systolic dysfunction.