Web-Based Application of Simplified Machine Learning for Detecting Reduced LVEF From 12-Lead ECG

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Deep learning (DL) models show promise in identifying reduced left ventricular ejection fraction (LVEF) from electrocardiograms (ECGs).
• The complexity of DL models hinders their widespread clinical adoption.
• Simplified machine learning (ML) models offer a potential solution for accessible LVEF detection.

Purpose of the Study:

• To develop and validate simplified ML models for detecting LVEF < 40% using 12-lead ECG numerical parameters.
• To create a user-friendly web application for implementing these ML models.

Main Methods:

• Retrospective analysis of ECG and echocardiography data from 21,471 patients across two institutions.
• Development and external validation cohorts were established, including patients with and without atrial fibrillation (AF).
• Four ML algorithms (random forest, XGBoost, support vector machine, generalized additive models) were evaluated for predicting continuous and binary LVEF outcomes.

Main Results:

• For continuous LVEF prediction, random forest (RF) showed moderate internal R-squared values (0.68-0.74) but poor external validation performance.
• For binary classification (LVEF < 40%), all models achieved high AUCs (>0.90) in the non-AF group internally.
• RF and XGBoost demonstrated strong performance in the AF group (AUC >0.90 internally) and adequate external validation accuracy (AUCs 0.80-0.90).

Conclusions:

• A simplified, web-based tool for preliminary screening of reduced LVEF using 12-lead ECG parameters has been successfully developed.
• These ML models offer a practical approach for early identification of patients with reduced LVEF.