An Ensemble Deep Learning Algorithm for Structural Heart Disease Screening Using Electrocardiographic Images: PRESENT...

Area of Science:

• Artificial Intelligence in Cardiology
• Deep Learning for Medical Imaging
• Electrocardiography Analysis

Background:

• Early identification of structural heart diseases (SHDs) is crucial for disease management.
• Cardiac imaging, essential for SHD diagnosis, faces accessibility limitations.

Purpose of the Study:

• To develop an automated deep learning approach for detecting and predicting multiple SHDs from 12-lead ECG images.
• To create a composite screening tool for SHDs using an ensemble deep learning model.

Main Methods:

• Developed convolutional neural network models to detect individual SHDs from ECG images, with SHD definitions based on transthoracic echocardiograms.
• Created an ensemble XGBoost model, PRESENT-SHD, for composite SHD screening.
• Validated PRESENT-SHD across multiple US hospitals, the ELSA-Brasil study, and UK Biobank for screening and risk stratification.

Main Results:

• PRESENT-SHD achieved an AUROC of 0.886 in the held-out test set, with 90% sensitivity and 66% specificity.
• The model demonstrated consistent performance across diverse demographic subgroups, ECG formats, and even smartphone-captured ECGs.
• A positive PRESENT-SHD screen indicated a 2- to 4-fold increased risk of new-onset SHD/heart failure, independent of other factors.

Conclusions:

• PRESENT-SHD is a validated AI-ECG tool capable of identifying various SHDs from 12-lead ECG images.
• This technology offers a robust, scalable, and accessible method for automated SHD screening and risk stratification.