A deep learning model could screen for coronary heart disease from a "pseudo-normal" electrocardiogram

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Coronary heart disease (CHD) screening is crucial, yet challenging in patients with normal or near-normal electrocardiograms (ECGs).
• Developing advanced methods for early CHD detection is essential for improving patient outcomes.

Purpose of the Study:

• To develop and validate a deep learning model (DLM) for rapid screening of coronary heart disease (CHD).
• To specifically assess the DLM's performance in identifying CHD among patients presenting with "pseudo-normal" ECGs.

Main Methods:

• Utilized standard 12-lead ECGs from CHD and non-CHD patients for DLM training and validation (SAH dataset).
• Employed an external testing set of ECGs from CHD patients who underwent revascularization (FAH dataset).
• Evaluated diagnostic performance using Area Under the Receiver Operating Characteristic Curve (AUC) and eigenvalue-based visual cluster analysis.

Main Results:

• The DLM achieved high AUC values: 0.913 for the internal SAH dataset and 0.936 for the external FAH dataset.
• Demonstrated a notable AUC of 0.721 for identifying CHD in patients with "pseudo-normal" ECGs.
• Identified specific DLM parameters with potential as significant indicators for screening "pseudo-normal" ECGs.

Conclusions:

• The developed DLM effectively facilitates rapid CHD screening via ECGs.
• The model shows particular efficacy in detecting CHD in patients with "pseudo-normal" ECGs, augmenting traditional diagnostic methods.