Deep Learning-Based Electrocardiogram Analysis Predicts Biventricular Dysfunction and Dilation in Congenital Heart...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Artificial intelligence-enhanced electrocardiogram (AI-ECG) analysis shows potential for identifying biventricular pathophysiology.
• However, the application of AI-ECG analysis in congenital heart disease (CHD) requires further investigation.

Purpose of the Study:

• To develop and externally validate an AI-ECG model.
• The model aims to predict cardiovascular magnetic resonance (CMR)-defined biventricular dysfunction and dilation in patients with CHD.

Main Methods:

• A convolutional neural network was trained and tested on paired ECG-CMR data from patients with and without CHD.
• The model was designed to detect left ventricular (LV) dysfunction, RV dysfunction, and LV/RV dilation.
• Performance was evaluated using area under the receiver-operating curve (AUROC) and area under the precision recall curve during internal testing and external validation.

Main Results:

• The model demonstrated consistent performance in both internal testing and external validation cohorts.
• AUROC values for LV dysfunction, LV dilation, RV dysfunction, and RV dilation were high across both cohorts.
• Performance was lowest in patients with functionally single ventricles, and specific ECG features like QRS widening and T-wave inversions were identified as high-risk indicators for RV dysfunction/dilation.

Conclusions:

• AI-ECG analysis holds significant promise for predicting biventricular dysfunction and dilation in CHD patients.
• This technology may assist in optimizing the timing of cardiovascular magnetic resonance (CMR) imaging in CHD management.