AI Outperforms Humans at Analyzing Long-Term ECG Recordings

This analysis is a labor-intensive process. Ambulatory ECGs must be examined by specially trained personnel, known as ECG technicians, but there is a global shortage of approximately 15 million healthcare workers. This shortage creates significant bottlenecks in healthcare, and at the same time, patients could benefit from more extensive and longer ambulatory ECG recordings rather than shorter ones. Now, a new study has shown that an artificial intelligence (AI) network algorithm designed to interpret ambulatory ECG signals can surpass humans in analyzing long-term ECG recordings.

The study conducted by researchers at Lund University (Lund, Sweden) and Population Health Research Institute (PHRI, Hamilton, ON, Canada) included 14,606 individual patients who recorded an average of 14 days of ECG data, totaling over 200,000 days of ECG data. These data were initially reviewed by ECG technicians using standard clinical methods. The same data were then re-analyzed using the DeepRhythmAI algorithm, developed by MEDICALgorithmics (Warsaw, Poland), specifically for this task. The results, published in Nature Medicine, demonstrated that the AI analysis resulted in 14 times fewer missed diagnoses of severe arrhythmias, such as complete heart block, ventricular tachycardia, and atrial fibrillation. The AI missed severe arrhythmias in 0.3% of patients, compared to 4.4% for the technicians.

The AI model was able to rule out severe arrhythmia with 99.9% confidence in a 14-day ECG recording. The number of false positives, where serious arrhythmias were misinterpreted, was relatively low—12 per 1,000 recording days for the AI compared to 5 per 1,000 recording days for human analysis. This study is the first to evaluate not just the AI’s ability to assess individual ECG strips, but also to explore what would happen if human technicians were replaced by AI. The researchers aimed not to compare AI’s performance with cardiologists in diagnosing specific arrhythmias, but to assess the potential impact of replacing technicians and having physicians receive reports directly from AI. If successful, this approach could represent a major innovation, helping to address the global shortage of trained staff for interpreting long-term ECG monitoring.

“Today, most long-term ECG devices use some type of AI to support interpretation, but with varying quality,” said Jeff Healey, senior scientist at the PHRI. “And there are still long waiting times for long-term ECG monitoring, in some cases many months. If we have a qualified AI model that can review all ECGs, then we would have both much cheaper and faster diagnostics.”

Related Links:
Lund University
PHRI
MEDICALgorithmics