Validation of an automated artificial intelligence system for 12‑lead ECG interpretation

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• The electrocardiogram (ECG) is a vital diagnostic tool for cardiac patients.
• Current computerized ECG interpretation (CIE) accuracy is less than that of physicians.
• This study assesses an AI-powered ECG system against state-of-the-art CIE.

Purpose of the Study:

• To evaluate the diagnostic performance of an AI-powered ECG system.
• To compare the AI system's accuracy with existing CIE methods.
• To assess the AI system's ability to detect 20 essential diagnostic patterns.

Main Methods:

• An AI system with 6 deep neural networks (DNNs) was trained on 12-lead ECGs.
• The system was trained to detect 20 diagnostic patterns across 6 categories.
• Performance was evaluated against a reference standard of expert cardiologist consensus.

Main Results:

• The AI system achieved high F1 scores across all 6 diagnostic categories (e.g., Rhythm 0.957, ACS 0.925).
• DNNs surpassed state-of-the-art CIE for 13 out of 20 diagnostic patterns.
• The AI system demonstrated non-inferior performance for remaining diagnoses.

Conclusions:

• The AI-powered ECG model accurately identifies electrocardiographic abnormalities.
• The AI system shows significant potential as a clinical tool for healthcare professionals.
• This technology can enhance the diagnostic capabilities of ECG interpretation.