1271 Evaluation of a Novel In-Ear Wearable for HRV and Microarousal Detection During Sleep

Abstract Introduction Heart rate variability (HRV) is a key biomarker of autonomic function and arousal during sleep. Traditional methods for HRV analysis rely on research-grade electrocardiography (ECG) or finger-based photoplethysmography (PPG), both of which pose limitations in comfort and signal stability during overnight wear. We present preliminary Phase I results from a novel in-ear wearable (Lumia) designed to enable continuous HRV tracking and microarousal detection during sleep. Methods Lumia is a lightweight, in-ear PPG sensor with integrated memory and onboard feature extraction. HR and HRV metrics were derived from Lumia’s PPG signal and benchmarked against research-grade ECG and finger PPG during overnight recordings. Separately, a microarousal detection algorithm was developed using research-grade finger PPG from 10 full-night polysomnography (PSG) datasets from unique subjects. A random forest model was trained using leave-one-out cross-validation and focused on changes in HR, RR interval variability, and pulse amplitude in pre- and post-arousal windows. Results Compared to research-grade finger PPG (MAE = 6.94; RMSE = 24.63 ms), Lumia (MAE = 4.25; RMSE = 18.55 ms) demonstrated significantly lower error in HRV estimation. Notably, the wearable’s resting-state HRV data during sleep was highly correlated with ECG-derived values, supporting feasibility for overnight use. In parallel, the initial microarousal model achieved an average cross-validated AUC of 0.90 across 10 PSG subjects. Predicted microarousal counts per subject showed strong correlation with ground truth (r = 0.902). Key predictive features included HR standard deviation and RR interval variability. Conclusion These findings support the technical feasibility of the Lumia wearable for high-fidelity HR and HRV monitoring during sleep and offer promising early evidence for microarousal detection using machine learning. The combination of stable in-ear PPG signals and embedded analytics highlights Lumia’s potential as a fieldable, real-time sleep monitoring solution. The strong correlation between predicted and actual microarousal counts suggests good generalizability. Lumia’s superior HRV accuracy over finger PPG indicates that model performance may improve further with in-ear data. Future work will focus on training the microarousal detection algorithm using labeled in-ear PPG data to further enhance performance and clinical utility. Support (if any) This research was supported by the Defense Health Agency, awarded to Lumia Health.