Reliability of pulse photoplethysmography sensors: Coverage using different setups and body locations

Abstract

Pulse photoplethysmography (PPG) is a simple and economical technique for obtaining cardiovascular information. In fact, PPG has become a very popular technology among wearable devices. However, the PPG signal is well-known to be very vulnerable to artifacts, and a good quality signal cannot be expected for most of the time in daily life. The percentage of time that a given measurement can be estimated (e.g., pulse rate) is denoted coverage (C), and it is highly dependent on the subject activity and on the configuration of the sensor, location, and stability of contact. This work aims to quantify the coverage of PPG sensors, using the simultaneously recorded electrocardiogram as a reference, with the PPG recorded at different places in the body and under different stress conditions. While many previous works analyzed the feasibility of PPG as a surrogate for heart rate variability analysis, there exists no previous work studying coverage to derive other cardiovascular indices. We report the coverage not only for estimating pulse rate (PR) but also for estimating pulse arrival time (PAT) and pulse amplitude variability (PAV). Three different datasets are analyzed for this purpose, consisting of a tilt-table test, an acute emotional stress test, and a heat stress test. The datasets include 19, 120, and 51 subjects, respectively, with PPG at the finger and at the forehead for the first two datasets and at the earlobe, in addition, for the latter. C ranges from 70% to 90% for estimating PR. Regarding the estimation of PAT, C ranges from 50% to 90%, and this is very dependent on the PPG sensor location, PPG quality, and the fiducial point (FP) chosen for the delineation of PPG. In fact, the delineation of the FP is critical in time for estimating derived series such as PAT due to the small dynamic range of these series. For the estimation of PAV, the C rates are between 70% and 90%. In general, lower C rates have been obtained for the PPG at the forehead. No difference in C has been observed between using PPG at the finger or at the earlobe. Then, the benefits of using either will depend on the application. However, different C rates are obtained using the same PPG signal, depending on the FP chosen for delineation. Lower C is reported when using the apex point of the PPG instead of the maximum flow velocity or the basal point, with a difference from 1% to even 10%. For further studies, each setup should first be analyzed and validated, taking the results and guidelines presented in this work into account, to study the feasibility of its recording devices with respect to each specific application.

BrainBeats, an Open-Source EEGLAB Plugin to Jointly Analyze EEG and Cardiovascular Signals

“…PPG signals are smoother and lack distinct features relative to ECG, making them vulnerable to artifacts 75 . While the algorithm used in this study was previously validated and performed well on the dataset used for this study, it might not perform as well on other types of PPG signals, especially those collected with wearable technologies.…”

A Novel PPG-Based Biometric Authentication System Using a Hybrid CVT-ConvMixer Architecture with Dense and Self-Attention Layers

“…It involves the use of light sensors to capture the reflected light from the tissue, providing insights into cardiovascular activity and physiological characteristics. PPG signals can be obtained from various parts of the body, including the fingertip, wrist, or earlobe [ 4 ]. These signals exhibit unique patterns that are specific to individuals, making them suitable for biometric authentication [ 5 ], as shown in Figure 1 .…”

“…It involves the use of light sensors to capture the reflected light from the tissue, providing insights into cardiovascular activity and physiological characteristics. PPG signals can be obtained from various parts of the body, including the fingertip, wrist, or earlobe [4].…”

“…A biometric authentication system utilizes the CVT-ConvMixer hybrid deep learning model, enhancing robustness and performance by effectively managing signal variations in PPG data for accurate individual identification. 4.…”