Signal Quality Index (SQI) in PPG Analysis
A Signal Quality Index (SQI) is a numerical metric that automatically quantifies the reliability of a PPG segment for physiological analysis. SQIs typically range from 0 (unacceptable) to 1 (clean signal) and are used to gate downstream algorithms — only high-quality segments are processed for SpO2, HRV, or other metrics.
Multiple SQI approaches have been developed, each targeting different artifact manifestations. Template-based SQI methods (e.g., Li et al., 2012) compute the cross-correlation between each detected beat and a reference template derived from the same recording. Correlation coefficients below 0.9 flag beats as potentially corrupted. Spectral SQI methods compare the power spectrum of the PPG segment with an idealized cardiac spectrum, penalizing excess power outside the expected heart rate harmonics. Morphological SQI methods check physiological plausibility of waveform features (amplitude, width, diastolic notch presence).
The PhysioNet Computing in Cardiology Challenge 2014 spurred extensive development of combined SQI frameworks. The SQIABP and SQIPPG indices from that challenge demonstrated that ensemble SQI approaches combining template, spectral, and statistical features outperform single-metric approaches, achieving AUC > 0.95 for artifact detection.
In clinical monitoring, SQI thresholds directly impact alarm fatigue. ICU monitors using physiologically-informed SQI gating can reduce false SpO2 alarms by 60–80% compared to amplitude-based detection alone. Consumer wearables typically use simpler SQI proxies — signal standard deviation, autocorrelation peak height, or accelerometer-based motion detection — to trigger measurement retry rather than continuous quality monitoring.
Frequently Asked Questions
What SQI threshold should be used before HRV analysis?
Most guidelines recommend excluding segments with SQI < 0.8 (template correlation-based) or with >5% beats flagged as artifacts before computing any HRV metrics.
Can machine learning improve SQI accuracy?
Yes. CNN and LSTM-based SQI classifiers trained on annotated clinical datasets achieve F1 scores of 0.93–0.97, outperforming handcrafted feature-based approaches in complex multi-artifact scenarios.
How does SQI differ from signal-to-noise ratio (SNR)?
SNR is a general measure of signal clarity relative to noise, typically computed in the frequency domain. SQI is a clinically-oriented composite metric that assesses physiological plausibility, not just spectral purity.