Development of Clinically Validated Artificial Intelligence Model for Detecting ST-segment Elevation Myocardial
Sang-Hyup Lee1, Kyu Lee Jeon2, Yong-Joon Lee1
1Division of Cardiology, Severance Hospital, Yonsei University College of Medicine, Seoul, South Korea.
Annals of Emergency Medicine
|July 27, 2024
Summary
An artificial intelligence (AI) model accurately diagnoses ST-segment elevation myocardial infarction (STEMI) using electrocardiogram (ECG) data. This AI tool aids in timely cardiac catheterization laboratory activation, improving patient care for STEMI.
Area of Science:
- Cardiology
- Artificial Intelligence
- Medical Diagnostics
Background:
- Primary percutaneous coronary intervention is crucial for ST-segment elevation myocardial infarction (STEMI).
- Cardiac catheterization laboratory activation for STEMI is often suboptimal.
- Accurate and timely diagnosis is essential for effective STEMI treatment.
Purpose of the Study:
- To develop a precise artificial intelligence (AI) model for diagnosing STEMI.
- To improve the accuracy of cardiac catheterization laboratory activation for STEMI patients.
Main Methods:
- Utilized electrocardiogram (ECG) waveform data from a Korean percutaneous coronary intervention registry.
- Developed a deep ensemble model combining 5 convolutional neural networks.
- Validated the AI model using clinical data, physician comparisons, and external datasets.
Main Results:
- The AI model achieved 92.1% accuracy, 95.4% sensitivity, and 91.8% specificity on 18,697 ECGs.
- Demonstrated outstanding and balanced performance across clinical validation, physician comparison, and external validation.
- The AI model exhibited reasonable explainability via gradient-weighted class activation mapping.
Conclusions:
- The deep ensemble AI model demonstrates robust performance in diagnosing STEMI.
- The AI model shows potential for improving cardiac catheterization laboratory activation efficiency.
- Further prospective validation is recommended to confirm clinical benefits in real-world settings.
