PulseAI Algorithm and Assessment Framework Sets New Standard for Wearable ECG Performance
A new peer-reviewed study published in Heart Rhythm demonstrates that PulseAI’s advanced artificial intelligence (AI) technology significantly improves the diagnostic accuracy and reliability of wearable ECG devices such as the Apple Watch.
The paper, “Accounting for Inconclusive Results and Repeated Testing: A Framework for Evaluating Wearable ECG Diagnostic Performance with Application to Apple Watch and AI-Enhanced Interpretation”, was authored by researchers from PulseAI Ltd, Ulster University, and clinical collaborators across Ireland, the UK and Germany.
This landmark work introduces a new way to evaluate the real-world performance of wearable ECGs, one that accounts for inconclusive readings and repeat testing, key factors often ignored in traditional validation studies.
Key Findings
✅ 92% reduction in inconclusive results
PulseAI’s neural network reduced inconclusive Apple Watch readings from 17.6% to just 1.4%, providing conclusive diagnostic feedback almost every time.
✅ Higher diagnostic performance
Under the realistic “intention-to-diagnose” framework, which includes repeat testing, the AI model achieved 98.4% sensitivity and 96.6% specificity for atrial fibrillation (AF) detection. The Apple Watch’s native algorithm reached 92.2% sensitivity and 91.0% specificity.
✅ Near-perfect repeatability
Across repeated tests, PulseAI achieved κ = 0.96, indicating almost perfect reliability.
By comparison, the Apple Watch algorithm alone achieved κ = 0.77, a level considered substantial but not ideal for clinical confidence.
Why This Matters
Most wearable ECG studies report inflated accuracy by excluding inconclusive results and assuming users take only one reading.
PulseAI’s research challenges this outdated practice by applying the intention-to-diagnose framework, capturing how people actually use wearable ECGs at home.
When this more realistic analysis was applied, previously reported diagnostic performance fell sharply, revealing that conventional methods can overstate accuracy by up to 15–20 percentage points.
“Our framework captures how people actually use wearable ECGs,” said Peter Doggart, CTO of PulseAI and lead author of the study. “By accounting for repeated testing and inconclusive results, we’re setting a new benchmark for transparency, reproducibility, and clinical trust in AI-enabled cardiac diagnostics.”
AI That Understands the Heart
The PulseAI algorithm was trained on over one million ECGs from diverse international datasets and designed to interpret a broader range of rhythms beyond atrial fibrillation, such as sinus rhythm with ectopy. This comprehensive training enables the system to provide more accurate, conclusive, and clinically useful interpretations without requiring changes to wearable hardware.
The study highlights that the challenge isn’t the quality of the Apple Watch signal, it’s the software interpreting it. PulseAI’s AI engine transforms existing consumer devices into high-performance diagnostic tools suitable for real-world clinical monitoring.
Figure 1. Comparison of performance metrics between PulseAI platform vs Apple algorithm for detection of atrial fibrillation
Clinical Impact
By sharply reducing inconclusive results and increasing repeatability, PulseAI’s technology:
Reduces patient anxiety caused by inconclusive readings
Improves clinician confidence and adoption of wearable ECG technology
Enhances the efficiency of remote patient monitoring
“These findings bring consumer health technology closer to clinical-grade performance,” said Dr. Alan Kennedy, CEO at PulseAI. “It’s a major step forward for early atrial fibrillation detection and stroke prevention.”
