Development and validation of a deep-learning algorithm for rule-in and rule-out coronary artery disease based on...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Current coronary artery disease (CAD) guidelines utilize pretest probability (PTP) thresholds (≤5% for rule-out, ≥15% for rule-in) for patient management.
• Electrocardiograms (ECGs) are a common diagnostic tool, but their utility in ruling out CAD without evidence of ischemia is an area for advancement.

Purpose of the Study:

• To develop and validate a deep-learning algorithm for ruling in or out coronary artery disease (CAD).
• To assess the algorithm's performance using electrocardiograms (ECGs) in patients without evidence of myocardial ischemia.
• To establish an algorithm-based decision-making protocol for CAD risk stratification.

Main Methods:

• A deep-learning algorithm was trained and internally validated on 90% and 10% of ECG data from Fuwai Hospital, respectively.
• The algorithm was designed to detect CAD (≥70% stenosis) using 12-lead ECGs.
• An external validation was performed using data from Yunnan Fuwai Hospital to assess the decision-making protocol's performance, aiming for a 95% negative predictive value (NPV).

Main Results:

• Internal validation showed an AUC of 0.81. Rule-out and rule-in groups had CAD prevalences of 5% and 23%, respectively.
• External validation demonstrated rule-out and rule-in CAD prevalences of 0% and 15%, respectively.
• The algorithm-based protocol achieved guideline-recommended performance in patient stratification.

Conclusions:

• The developed deep-learning algorithm effectively detects CAD using ECGs, even without signs of myocardial ischemia.
• An algorithm-based decision-making protocol can successfully guide the rule-out or rule-in process for further patient examination.
• This AI approach shows promise in improving the diagnostic pathway for suspected coronary artery disease.