Automatic triage of twelve-lead electrocardiograms using deep convolutional neural networks: a first implementation...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Expert interpretation of electrocardiograms (ECGs) is often limited.
• An artificial intelligence (AI)-based triage algorithm, DELTAnet, can assist physicians in prioritizing ECGs.
• Clinical implementation of AI for ECG triage is understudied.

Purpose of the Study:

• To prospectively evaluate the clinical implementation and impact of the DELTAnet AI algorithm for ECG triage.
• To assess the performance of DELTAnet in classifying ECGs compared to final clinical diagnoses.
• To investigate the association between DELTAnet's predicted triage classes and patient clinical outcomes.

Main Methods:

• Prospective cohort study involving adult patients in non-cardiology departments undergoing routine ECGs.
• DELTAnet classified 12-lead ECGs into triage categories: normal, abnormal not acute, subacute, and acute.
• Performance was compared against triage based on final clinical diagnosis; clinical outcomes were analyzed.

Main Results:

• 1061 patients and ECGs were analyzed.
• DELTAnet demonstrated high performance with a mean concordance statistic of 0.96 (95% CI 0.95-0.97) compared to clinical triage.
• No critical ECGs requiring intervention were missed, and false positives for acute triage were minimal (2.6%).

Conclusions:

• This is the first prospective study on the clinical implementation of an AI-based ECG triage algorithm.
• DELTAnet is effective and safe for clinical use in triaging 12-lead ECGs within non-cardiology hospital settings.
• The AI algorithm supports efficient ECG prioritization, potentially improving time to treatment for acute cardiac conditions.