HealthRing: Physiology Dataset for Health Sensing on Rings

Smart rings enable unobtrusive monitoring of cardiovascular vital signs via photoplethysmography (PPG), yet rigorous validation is limited by the scarcity of open, multi-parameter datasets. We present HealthRing, three synchronized cohorts from 54 adults: (i) scripted physiological stimuli (Study 1), (ii) semi-free-living daily activities (Study 2), and (iii) motion-intensive treadmill running with natural arm swing that induces substantial motion artefacts (Study 3). Two custom rings-reflective and transmissive-record infrared/red PPG and 3-axis accelerometer data at 100 Hz, time-aligned to clinical-grade heart rate, respiratory rate, peripheral oxygen saturation, and blood pressure. On the controlled+daily cohorts (Study 1/2), physics-based and supervised benchmarks reach mean absolute errors of 5.33 BPM (HR), 2.98 breaths/min (RR), 1.72% (SpO₂), 12.98 mmHg (SBP), and 7.64 mmHg (DBP). On the treadmill cohort, fine-tuning cuts HR error from 36.91 to 23.99 BPM and RR error from 5.44 to 4.61 breaths/min relative to zero-shot transfer, stress-testing motion robustness. A publicly available toolkit (RingTool) provides preprocessing, classical signal processing, and deep learning pipelines. HealthRing closes critical gaps in ring-based cardiovascular sensing and supports algorithm development across controlled labs, daily life, and in-the-wild running.