Physiology you can trust: tracking change between visits
Between two consultations, is a patient getting better or quietly getting worse? Wearables and phones stream heart rhythm, sleep and movement around the clock; the hard part is trusting them and telling stress from a coffee, a run or a bad night. The team will turn these everyday signals into a measure of change from each person's own baseline, decide who should hear about a meaningful shift and when, and keep alert fatigue out of the loop. Open wearable datasets, foundation models for physiological signals and language models are on the table, along with the outline of a first study with patients. One lead worth chasing: the body clock as an early signal.
Resources
- Wearables and phone sensing
- Sensitive data handling
- WESAD ↗
- Wearable stress and structured exercise sessions (PhysioNet, ODC-By) ↗
- MMASH — 24 h heart signals, actigraphy, sleep, cortisol and melatonin (PhysioNet, ODbL) ↗
- LifeSnaps — 4 months of Fitbit data and surveys (Zenodo, CC BY 4.0) ↗
- PMData — 5 months of Fitbit data and self-reports (Simula) ↗
- BUT PPG — smartphone PPG with quality labels (PhysioNet, CC BY 4.0) ↗
- PPG-DaLiA — wrist PPG with motion during daily activities (UCI) ↗
- Wearable exam stress — real-world stress (PhysioNet, ODC-By) ↗
- Nurse stress — wristband data on hospital shifts (Dryad, CC0) ↗
- Autonomic Aging — HRV across ages incl. older adults (PhysioNet, ODbL) ↗
- PTB-XL — 12-lead ECG database (PhysioNet) ↗
- GLOBEM — multi-year phone and wearable data with weekly depression surveys (PhysioNet, credentialed access) ↗
- Speech models
- Data platform and governance
- Lab equipment
- Video / vision models
- Language-model endpoints
- Polar H10 straps ↗
- NeuroKit2 — physiological signal processing and quality ↗
- HeartPy — heart-rate analysis for PPG and ECG ↗
- PaPaGei — open foundation models for optical physiological signals (BSD-3) ↗
- pyVHR — camera-based heart rate (GPL-3.0) ↗
