Deep learning model for identifying significant tricuspid regurgitation using standard 12-lead electrocardiogram

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Transthoracic echocardiography (TTE) is the standard for detecting tricuspid regurgitation (TR) but is costly and operator-dependent.
• The 12-lead electrocardiogram (ECG) is readily available, presenting an opportunity for alternative diagnostic methods.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for detecting significant TR using 12-lead ECG signals and clinical data.
• To assess the diagnostic performance of DL models in identifying TR compared to traditional methods.

Main Methods:

• A dataset of 5432 patients with both ECG and TTE data was used, with 570 diagnosed with significant TR.
• A DL model architecture combining 1D convolutional neural networks, efficient channel attention, and Multihead Attention modules was employed.
• The model was trained on 3910 patients and validated on internal and external cohorts.

Main Results:

• The DL model achieved an accuracy of 0.762 and an AUC of 0.857 for predicting significant TR using ECG signals, age, and sex.
• Incorporating additional clinical factors improved sensitivity to 0.836 and AUC to 0.866, with satisfactory external validation.
• The DL model demonstrated substantial diagnostic performance in identifying significant TR.

Conclusions:

• Deep learning models utilizing ECG data show promise for facilitating the diagnosis of significant TR.
• This AI-driven approach could offer a more accessible and cost-effective screening tool for TR.
• Further clinical validation is necessary to establish the widespread utility of this DL model.