Unmixing the Neck: Accurate Jugular Venous Pulse Detection From Wearable PPG

Non-invasive monitoring of the jugular venous pulse (JVP) supports heart-failure management, but neck photoplethysmography (PPG) suffers from source ambiguity, in which arterial pulsations contaminate the venous waveform. We present a dual-channel wearable framework that automatically separates jugular venous from arterial waveforms.

The pipeline pairs a derivative-enhanced dynamic time warping (deDTW) algorithm with a stability-adjusted, distance-based feature-ranking scheme. deDTW quantifies the morphological and timing distance between candidate beats using a signal-quality-adaptive weighting of amplitude and derivative information. From dual-channel neck recordings of 15 volunteers we extracted 73 candidate features and trained a $k$-nearest-neighbor classifier under leave-one-subject-out (LOSO) cross-validation.

Optimization selected 28 features at $k=19$, yielding a classification accuracy of 98.65%. When combined with a nearest-neighbor distance-ratio rejection rule ($\tau = 1.005$), the system reached 99.82% sensitivity for arterial beats and a 98.73% F1-score for JVP detection, and rejected 93.88% of out-of-distribution ambiguous beats while retaining 99.74% precision for valid JVP signals.

The algorithm resolves venous-arterial source ambiguity, addressing a key limitation of wearable JVP analysis.

By acting as an automated source-verification gate, the method reduces inter-observer variability and supports continuous, non-invasive JVP monitoring for remote cardiovascular assessment. Validation here is in a healthy cohort; evaluation in patients with cardiovascular disease is the subject of future work.