Signal processing
Motion Artifact Detection in PPG
Motion artifact detection algorithms automatically identify PPG segments corrupted by physical movement using accelerometer-based activity detection, PPG signal statistics (amplitude excursions, spectral distortion), and machine learning classifiers, enabling quality-gated processing that improves measurement accuracy.
How it works
Schematic summary of this guide. Waveforms are illustrative, not measured data.
Motion artifacts in PPG arise from three mechanisms: sensor-skin interface displacement (dominant source), blood volume redistribution from tissue compression, and venous blood movement from inertial effects.
Accelerometer-based detection computes motion intensity metrics (root-mean-square acceleration, jerk magnitude, frequency content) and flags PPG segments where motion exceeds empirically determined thresholds. More sophisticated approaches compute the cross-correlation between accelerometer spectral content and PPG spectral content — high correlation indicates motion contamination of the PPG.
CNN classifiers processing 2–5 second PPG windows with simultaneous accelerometer input learn complex artifact patterns including slow drift, high-frequency oscillations, and motion-cardiac frequency overlap. These classifiers can operate at beat level (flag individual corrupted beats) or epoch level (flag 5–30 second segments).
Frequently Asked Questions
What accelerometer metric best predicts PPG artifact severity?
Higher-frequency jerk (derivative of acceleration) specifically targets the abrupt movements that cause the largest PPG artifacts.
Can motion artifacts be detected without an accelerometer?
Yes. PPG-only artifact detection uses signal statistics: abnormal amplitude excursions (>3× running IQR), loss of autocorrelation periodicity, spectral flatness increase, and template correlation decrease.
How should detected motion artifacts be handled?
Options include: exclusion (remove segment, simplest), interpolation (fill gaps with estimated values), and artifact removal (adaptive filtering, ICA). For clinical applications, exclusion with measurement unavailability flagging is safest. For consumer applications, interpolation maintains continuous display.