Atrial fibrillation detection in ambulatory patients using a smart ring powered by deep learning analysis of continuous photoplethysmography monitoring

Background Atrial fibrillation (AF) detection could be effective with photoplethysmography (PPG) signal monitoring by a wearable device. Purpose We aimed to validate the performance of AF detection among ambulatory patients who underwent electrical cardioversion for AF using a smart ring capable of continuous PPG monitoring and deep learning analysis. Methods In this prospective single-arm study, participants who underwent successful electrical cardioversion for AF were enrolled. The participants equipped a smart ring (CardioTracker, Sky Labs Inc., Seongnam, Republic of Korea) after the electrical cardioversion. The smart ring then continuously monitored PPG over 14 days to detect AF recurrence. The smart ring alarmed AF episodes based on deep learning analysis of PPG. The participants were asked to measure at least three daily ECGs using the smart ring to validate AF recurrence detected by PPG. All ECG snapshots were recorded along with lead I and saved with simultaneous PPG. ECG data were examined by the three cardiologists independently (SK, SRL, and EKC). The monitoring time, analyzable proportions of monitored signals, detection rates of AF episodes, and the diagnostic performance of PPG-based deep learning were evaluated. At the end of the monitoring, a survey on the use of the smart ring was performed. Results A total of 35 participants (mean age 58.9 years, male 74.3%) were enrolled. Figure 1 illustrates an example of PPG monitoring and PPG-ECG snapshots by the smart ring. The study participation period was a median of 14 days and the wearing time of the smart ring was a median of 9.2 days (IQR 7.1–11.5 days). Signal artifacts during daily activity decreased the analyzable proportions of monitored PPG by 68.5%. Irregular pulse episodes were detected by the smart ring in 29 (82.9%) participants after a median of 1 day from the cardioversion (Figure 2). A total of 2532 PPG-ECG snapshots were acquired and 1623 (64.1%) were interpretable by both the cardiologists (using ECG) and the deep learning analysis (using PPG). Comparing PPG by simultaneous ECG, the performance of AF detection by the smart ring was 98.7% for sensitivity, 97.8% for specificity, 2.2% for false positives, and 1.3% for false negatives (Figure 2). After using the smart ring, 76.9% of the participants responded that they had no discomfort in using the smart ring in daily activity and another 76.9% responded that it was helpful to monitor their disease. Conclusion Despite the signal artifacts during daily activity, AF detection with PPG monitoring by a smart ring could be effective for AF screening among ambulatory patients. Funding Acknowledgement Type of funding sources: Private company. Main funding source(s): This work was supported by Sky Labs Inc, Seongnam, Republic of Korea, and by the grant No. 0320202040 from the Seoul National University Hospital Research Fund.