Cardiac arrhythmias (CAs) are a leading cause of mortality worldwide. Continuous and long-term monitoring of at-risk populations is crucial to address their increasing prevalence, associated costs, and comorbidities. While traditional ambulatory electrocardiograms (ECG) have limitations, wearable devices using photoplethysmography (PPG) offer a unique opportunity for early detection of CAs, facilitating timely intervention and preventive care.
Most research studies and consumer devices primarily focus on detecting of atrial fibrillation (AF), the most common CA. However, other CAs often go undetected or impair the performance of AF detectors. Moreover, detecting CAs in free-living conditions presents greater challenges than in clinical settings. This thesis aims to tackle CA monitoring comprehensively by addressing the detection of various types of CA over extended recording periods in free-living conditions.
A comprehensive database of PPG recordings was created during this thesis, including long-term ambulatory recordings and various CAs . The reliability of several heartbeat detectors was then assessed in the presence of CAs, an essential step allowing for the extraction of numerous PPG features to characterize the spectral and the rhythmic properties of PPG signals as well as the morphology of individual PPG pulses. This foundation allowed for the development of classification models: binary models detecting abnormal cardiac rhythms, and multiclass models classifying PPG signals into several CA groups. All models were evaluated using long-term ambulatory recordings.
This manuscript introduces the reader to cardiac electrophysiology and PPG technology, followed by a description of the conducted clinical studies, and applied signal processing methods. Classification models and results are then presented, followed by an outline of the thesis limitations, and proposals for future improvements and research directions.
By exploring the hemodynamic signature of various CAs at the wrist, this thesis highlights both the potential and limitations of PPG-based wearable devices, paving the way for more robust and convenient CA monitoring in everyday life.