Each wearable sensor streams a noisy vital-sign reading to an on-device AI model — a digital twin of the patient that fuses every stream into one early-warning probability.
z = (HR - 70) / 15
raw = w1*z + w2*noise - w3*ln(n)
R = sigmoid(raw) * 100%
alert when R > threshold
- Heart rate — the chest sensor's live reading; deviation from the 70 bpm resting baseline (z) is the strongest risk signal.
- Sensor noise — measurement jitter from motion artefacts and skin contact; higher noise pushes the raw score up and around.
- Active sensors — more independent wearables (wrist, forehead, ankle…) reduce uncertainty roughly as 1/√n — this is sensor fusion, the same idea behind AI-assisted early-warning scores in hospitals.
- Alert threshold — the probability cutoff above which the digital twin raises an early warning, similar to how AI cancer-screening or deterioration models flag a case for human review.
This is the mechanism behind wearable AI monitoring and early-warning systems: no single sensor is trusted alone, the model's confidence grows with more corroborating data, and the alert fires on the fused probability, not a raw threshold on one signal.