Tracking with Wearables

Most autonomic tests separate the two branches of the nervous system by measuring heart activity and breathing at the same time. That's because the parasympathetic (vagal) signal rides on top of the baseline heart rhythm, and the sympathetic signal sits underneath it, so normally you need two dials to tell them apart. Our problem was: wearables usually only give you one dial — heart rate.

The breakthrough is a foundation model trained on years of paired, "fully-instrumented" autonomic tests. Those legacy tests showed us, frame by frame, where the parasympathetic band (what we call RFa, the respiratory-frequency area) really was, and therefore where the sympathetic band (LFa, the low-frequency area) had to be, because in autonomic physiology the parasympathetic boundary defines the sympathetic boundary. We fed thousands of these labeled examples into a model with enough nonlinear layers to "listen" into the spectrum of the heart and learn what vagal tone looks like — or, more accurately, what it "sounds" like — even when respiration isn't explicitly provided. Think of it like source-separating the oboe line from a full orchestra recording: if you've heard enough orchestrations, you can start pulling the oboe out of a single audio track.

Because of that training, the model can now accept only a reliable heart-rate stream from a commodity wearable and still output two signals: an estimated RFa (parasympathetic span) and the implied LFa (sympathetic span). In practice, this gets us "about 90% of the way there" for most people — meaning the trend and magnitude are close enough for clinical interpretation — and the remaining 10% are the complex cases whose signals are very weak, non-stationary, or distorted by disease, and who may still benefit from a full, in-person autonomic workup. This is exactly the right tradeoff for scaling: common cases get lightweight, frequent, wearable-based monitoring; rare cases get escalated.

The important part: this removes the biggest barrier to population-scale autonomic monitoring. We no longer have to wait months between tests or ship two sensors. A single heart-rate device becomes a window into both branches of the ANS, and the model handles the unmixing step for you.