AUREON is a browser-based tool for real-time acquisition, exploration, and analysis of beat-to-beat cardiac dynamics. It acquires RR intervals via Bluetooth LE from a Polar H10 chest strap, visualises the RR trajectory in a 3D phase-space representation (RRₙ × ΔRRₙ × physiological time), and computes a panel of validated heart-rate variability (HRV) metrics. All processing runs locally in the browser — no data are transmitted to any server.
RR intervals are parsed from the Bluetooth LE Heart Rate Measurement characteristic (0x2A37). Timestamps are browser host-clock timestamps (ms since epoch). Web Bluetooth requires a secure context (HTTPS or localhost) and is currently supported in desktop Chrome and Edge. The artifact filter removes beats outside 250–2000 ms and suppresses successive differences > 300 ms; it can be toggled in the Explore panel and is on by default.
All algorithms are implemented directly from primary literature. Reference values are validated against a Python/numpy reference implementation (seed = 42); the self-test runs automatically on page load and is inspectable in the browser console via aureonSelfTest().
| Parameter | Formula / Method | Primary Reference |
|---|---|---|
| SDNN | Sample SD of all NN intervals (ddof = 1).√( Σ(RRᵢ − R̄R)² / (N−1) ) |
Task Force of ESC/NASPE (1996). Eur Heart J 17:354–381. |
| RMSSD | Root mean square of successive differences.√( Σ(RRᵢ₊₁ − RRᵢ)² / (N−1) ) |
Task Force of ESC/NASPE (1996). Eur Heart J 17:354–381. |
| SD1 | Poincaré plot minor axis SD. Direct geometric computation from rotated coordinates: std( (RRᵢ₊₁ − RRᵢ) / √2 , ddof=1 ). Note: the algebraic shortcut SD1 = RMSSD/√2 holds only asymptotically and is not used. |
Brennan M et al. (2001). IEEE Trans Biomed Eng 48(11):1342–1347, Eq. 7. Hoshi RA et al. (2013). Front Physiol 4:175. |
| SD2 | Poincaré plot major axis SD. Direct geometric computation: std( (RRᵢ₊₁ + RRᵢ) / √2 , ddof=1 ). Note: the algebraic shortcut SD2 = √(2·SDNN² − SD1²) requires consistent sample bases and is not used. |
Brennan M et al. (2001). IEEE Trans Biomed Eng 48(11):1342–1347, Eq. 8. |
| DFA α1 | Detrended Fluctuation Analysis short-range scaling exponent. Scale range n = 4–16. F(n) = √(mean of all squared detrended residuals across segments at scale n). α1 is the slope of log F(n) vs log n. | Peng CK et al. (1995). Chaos 5(1):82–87. Makikallio TH et al. (1999). Am J Cardiol 83(6):836–839. Gronwald T et al. (2019). Front Physiol 10:1143. |
| SampEn | Sample Entropy. SampEn(m=2, r=0.2). Template length m = 2, tolerance r = 0.2 × std(RR). Chebyshev distance. Self-matches excluded. SampEn = −ln(A/B) where B = matches of length m, A = matches of length m+1. Returns NaN if N > 1000 (computational limit) or if A = 0. | Richman JS & Moorman JR (2000). Am J Physiol Heart Circ Physiol 278:H2039–H2049. Lake DE et al. (2002). Am J Physiol Regul Integr Comp Physiol 283:R789–R797. |
The main plot shows each heartbeat as a point in three-dimensional space: RRₙ (interval duration), ΔRRₙ = RRₙ − RRₙ₋₁ (beat-to-beat change), and physiological time τ (cumulative RR sum in seconds). This representation makes autonomic trajectory patterns — loops, clusters, and regime shifts — directly visible as geometric structures.
All data are processed locally in your browser. No RR data, session data, or metadata are transmitted to any external server. Avoid entering patient-identifiable information unless your local governance and ethics framework explicitly permits it.
Developed by Titus Keller, Neonatologist & Pediatric Intensivist. For academic collaboration enquiries, use the contact form at tituskeller.com.
DFA α1 boundaries are exploratory attention thresholds, not validated clinical alarm limits. Use clinical context and signal quality before interpretation.