MettleSense Shield · AI-Powered ECG Intelligence · Research Beta v2.0
An independently developed ECG intelligence platform combining deterministic signal processing, HRV analytics and machine learning to screen for patterns associated with multiple cardiac rhythm and ECG abnormalities from wearable and 12-lead ECG data.
Shield evaluates nine rhythm and ECG conditions through a defined sequential decision framework, producing screening outputs intended to support further clinical evaluation rather than diagnosis. A condition is only cleared after every prior condition is systematically evaluated.
HR >150 · CV <0.05 · Mandatory first check
Polar "??? ms" fingerprint · RMSSD >200 + HR >100 · CV >0.40
RR pause >3000 ms · Golden Rule: HR >20 cannot be asystole
HR 250–350 · CV <0.03 · OR · HR 120–160 · CV <0.02
HR >150 · CV <0.10 · Not VTach · Not AFlutter
CV >0.15 · SampEn >0.5 · ≥10 RR intervals
QTc >500 ms · Bazett formula + Boudoulas QT estimate
Athlete: <35 bpm · General: <60 bpm or >100 bpm
Normal sinus rhythm pattern — all 8 conditions evaluated without abnormalities
Shield computes all five. SampEn provides an additional non-linear measure of rhythm irregularity that complements conventional HRV metrics in AFib-pattern analysis, calibrated to athlete baselines rather than generic population norms.
Beat-to-beat parasympathetic tone. Your primary recovery readout. The metric every wearable shows — but only the beginning.
Total autonomic load. Overall HRV picture — sympathetic and parasympathetic combined.
Parasympathetic activity percentage. The easiest metric to interpret — percentage of beat pairs differing by >50 ms.
Rhythm irregularity index. AFib and AFlutter leave distinct signatures here that RMSSD alone misses.
Non-linear chaos index (m=2, r=0.2·SDNN). Adds a deeper measure of rhythm irregularity beyond conventional time-domain HRV metrics, strengthening AFib-pattern analysis alongside CV(RR).
Pan-Tompkins is the gold-standard 5-stage R-peak detection algorithm used in clinical ECG systems. Shield runs it entirely client-side — from raw Polar H10 CSV or Apple Watch export. No server. No API. 100% of your cardiac data stays on your device.
HR 51 bpm · RMSSD 47.8 ms · Normal Sinus Rhythm. Every metric live from the uploaded session.
Recovery score runs 0–100 when cardiac status is clear. The moment any arrhythmia flag fires — Shield suppresses recovery to N/A, prioritizing safety and ensuring that attention remains focused on the primary rhythm findings.
Workout recommendations — Running, HIIT, Cardio, Strength — appear only when the Athlete profile is active, all 9 classifier checks are clear, and recovery is computed and valid.
HR, RMSSD, SDNN, pNN50, CV(RR), SampEn, Recovery score, rhythm classification, dual-source confirmation, ICD-10 alert — all computed from a single uploaded PDF. No manual entry.
| Device | Format | What Shield extracts |
|---|---|---|
| Polar H10 | PDF (ECG & HRV Insights app) | BPM + RMSSD per page · "??? ms" VFib fingerprint |
| Apple Watch | PDF (30-sec ECG strips) | BPM + device classification + HR=0 auto-correction |
| Any ECG device | Raw CSV (Polar Sensor Logger) | Pan-Tompkins R-peak detection → real RR intervals |
Holter-capable: upload multiple sessions. Shield tracks RMSSD trend and athlete baseline over time.
A resting HR of 38 bpm is normal for an endurance athlete. Shield won't flag it as pathological. Toggle Athlete / General in the header — watch thresholds shift in real time.
| Case | Input | Expected | Result |
|---|---|---|---|
| VFib | Synthetic CSV | VFib flag + recovery suppressed | PASS |
| AFib | Polar H10 PDF | AFib + dual confirmation | PASS |
| SVT / VTach | Apple Watch PDF | SVT or VTach + dual confirmation | PASS |
| Athlete Sinus | Real Apple Watch PDF | Normal sinus rhythm + Recovery ≥70 | PASS |
Interested in this research, technical collaboration, or exploring the potential of this project? Feel free to get in touch.
research@mettlesense.com