Artificial Intelligence in Wearable ECG and PPG Devices for Arrhythmia Detection: A Narrative Review

Abstract

Cardiovascular disease is the world’s top cause of death. Short recording times and a dependence on clinical settings are two practical drawbacks of traditional methods like Holter monitoring and standard 12-lead ECG. Two significant trends have come together in recent years: the broad availability of consumer-grade wearable technology, including smartwatches and adhesive ambulatory ECG patches, and the quick advancement of deep learning and artificial intelligence. Four main themes currently dominate research in this field: diagnostic accuracy under controlled conditions, algorithmic innovation for real-time and on-device analysis, real-world implementation and performance in large prospective studies, and the persistent technical, clinical, and regulatory challenges that limit widespread deployment.AI-based systems have shown consistently high performance in controlled environments and curated datasets, with reported sensitivities and specificities commonly ranging between 92% and 99%.However, performance, especially specificity and generalizability, frequently declines significantly if these algorithms are exposed to the messiness of daily movement, a variety of patient demographics, fluctuating skin contact, and hardware limitations outside of the lab. More advanced transformer architectures and multimodal techniques that combine ECG signals with photoplethysmography (PPG) waveforms, as well as lightweight convolutional neural networks that can operate directly on the device itself and provide true real-time detection, have been the focus of recent algorithmic work. The present status of wearable ECG and PPG technology for AI-enabled arrhythmia detection is examined in this review. It critically evaluates diagnostic performance, translational gaps that still exist and paying  attention to the  prospective studies that have shaped the field. All things considered, the data indicates that wearables driven by AI have real potential for screening for arrhythmias at the population level and have already shown promising signs of clinical utility. The field will require larger, longer, and more diversified prospective trials that are sufficiently powered to identify significant differences in hard clinical outcomes.