Correlating Electrocardiograms with Echocardiographic Parameters in Hemodynamically-Significant Aortic Regurgitation...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Aortic regurgitation (AR) presents challenges for deep learning (DL) detection in electrocardiograms (ECGs).
• The extent to which ECGs can detect AR-related left ventricular (LV) remodeling and dysfunction remains unclear.

Purpose of the Study:

• To evaluate the capability of DL-based ECG models in predicting LV remodeling parameters associated with hemodynamically significant AR.
• To assess the diagnostic performance of DL models for key indicators of LV dysfunction and remodeling.

Main Methods:

• Retrospective collection of 1457 12-lead ECGs from 573 patients with moderate-severe AR before aortic valve surgery.
• A ResNet-based DL model was developed to predict LV ejection fraction, dimensions, mass, and volume indices from ECGs.
• Model performance was evaluated using five-fold cross-validation and a held-out testing set.

Main Results:

• The DL model demonstrated significant predictive capabilities for various LV remodeling parameters, including LVEF (AUROCs 0.77-0.87) and dimensions (AUROCs 0.74-0.85).
• The model achieved notable AUROCs for LVEDD (>65 mm), LVESDi (>30 mm/m²), LVESVi (>45 ml/m²), LVEDVi (>99 ml/m²), LVMi (>158 mm/m²), and BAV (0.74-0.85).
• SHapley Additive exPlanations indicated the model primarily utilized the QRS complex for its predictions.

Conclusions:

• DL models can identify correlations between ECGs and LV remodeling/dysfunction in patients with significant AR.
• ECG analysis using DL may facilitate early detection of LV dysfunction and guide the need for echocardiography.
• This approach is particularly valuable in settings where timely echocardiographic assessment may be challenging.