Evaluation of the Signal Quality of Wrist-Based Photoplethysmography

Abstract

Detection of paroxysmal atrial fibrillation requires continuous cardiovascular monitoring due to its episodic nature. Such monitoring is impractical with electrocardiogram Holter monitors, which are the currently employed for ambulatory cardiovascular monitoring, but are cumbersome for prolonged use. This thesis studies monitoring using photoplethysmography (PPG) devices, which may be embedded into wristband devices which can be easily worn continuously. However, the quality of wrist-based PPG is highly variable, and is subject to artifacts from motion and other interferences.The goal of this thesis is to evaluate the signal quality obtained from wrist-based PPG when used in an ambulatory setting. Ambulatory data is collected over a 24-hour period for 10 elderly, and 16 non-elderly participants. Visual assessment is used as the gold standard for PPG signal quality, with Fleiss's Kappa being used to evaluate the agreement between raters. With this gold standard, 5 classifiers are evaluated using a modified 13-fold cross-validation approach. Based on this evaluation, a Random Forest quality classification algorithm is selected, with an accuracy of 74.5%. The algorithm is used to evaluate the ambulatory use of wrist-based PPG over a 24-hour period. Overall, it is found that data quality is high at night, and low during the day.

Evaluation of the signal quality of wrist-based photoplethysmography

“…The features used were the values at the first two peaks in the correlogram. These features were inspired by Wander and Morris (2014), and previously used in preliminary work done by our research team (Pradhan 2017). • Median noise ratio per pulse: This feature is the median of the signal-to-noise ratio (SNR) computed from each pulse in the PPG segment.…”

Photoplethysmography: A measure for the function of the autonomic nervous system in focal impaired awareness seizures

“…The use of PPG is limited because of its high sensitivity to motion. One study assessed the quality of the PPG signals using the Empatica E4 wristband and found that around 35% of data are considered high quality, that is, having “all identifiable pulses,” and this quality improves appreciably during sleep, when motion is less likely 20 . PPG signals may also be affected by orthostatic changes 34 .…”

“…Nevertheless, PPG‐based heart rate from wrist‐worn devices can be reliably used for monitoring when hand movements are limited, 35 and signals can be further cleaned to remove artifacts and improve cardiovascular monitoring 20,36 . Our samples included patients mostly in awake state (69%), and >95% of our signals had good quality (62% had fully identifiable pulses on the PPG signal, and another 33% had identifiable pulses for >75% of the portions 20 ). More than 69% of the signals had a completely flat accelerometry signal with no major orthostatic variations, as patients were in bed for the majority of the recording.…”

“…Because PPG signals are sensitive to motion, 1 we did not include any segments of the signals that were affected by motion artifacts as determined by a review of the PPG and three‐axis accelerometry signals. Therefore, we disregarded any signal portions that showed movement on the accelerometry signal and artifactual PPG waveforms 20 …”