Ambulatory detection of cardiac arrhythmias with wearable PPG sensors

Introduction

Cardiac arrhythmias contribute significantly to cardiovascular illness and mortality, yet they are frequently undiagnosed due to their intermittent or asymptomatic nature. Photoplethysmography (PPG) offers strong promise as a continuous, long-duration modality for identifying irregular heart rhythms. In this work, we introduce a new algorithm designed to detect abnormal rhythms using PPG combined with accelerometer data collected in everyday, ambulatory conditions.

Materials and Methods

ECG, PPG, and accelerometer recordings were obtained from 445 individuals referred for ambulatory ECG monitoring, divided into two study cohorts: a 24-hour cohort (45 participants) and a 7-day cohort (400 participants). Cardiac rhythms were automatically labelled using ECG-based software, and a cardiologist validated these annotations. The considered heart rhythms were normal sinus rhythm, sinus bradycardia, (sinus and supraventricular) tachycardia, and abnormal rhythms (atrial fibrillation and ventricular bigeminy).

We developed an algorithm that classifies rhythms into five classes based solely on PPG and accelerometer signals: normal, bradycardia, tachycardia, abnormal, and undecidable (representing low signal quality, motion artifacts, or uncertainty). The algorithm first extracts interbeat intervals and then classifies these interbeat intervals over 2-minute episodes into one of the classes mentioned above. Performance was assessed by comparing the predicted class of each episode to the corresponding ECG annotation.

Results

Table 1 summarizes the performance obtained in both study cohorts. For the 24-hour cohort, the algorithm reached an accuracy of 94.1% for abnormal rhythm detection, which further improved to 96.4% in the 7-day cohort. Atrial fibrillation accounted for most abnormal episodes (92.4% in the 24-hour cohort and 99.0% in the 7-day cohort). Both sensitivity and specificity were high and consistent across the two cohorts.

Conclusion

We present an algorithm capable of accurately identifying arrhythmias from PPG and accelerometer data in real-world ambulatory environments, achieving excellent specificity. This method shows strong potential for scalable implementation as a screening tool in populations with elevated arrhythmia risk.Table 1:Performance metrics.