Key result
Machine-learning analysis of reflectance-mode wrist photoplethysmography estimated total hemoglobin concentration with a mean absolute error of 12.6±1.7 g/L and correlation coefficient of 0.66±0.09.
Why the study?
Standard transmittance-mode PPG analysis methods do not apply to reflectance-mode sensors used in wrist wearables, requiring new approaches to noninvasively estimate total hemoglobin concentration.
Can machine-learning algorithms analyzing reflectance-mode MW-PPG signals from wrist-wearable devices accurately estimate total hemoglobin concentration compared to reference devices in volunteers?
Observational (n=170)
Can machine-learning algorithms analyzing reflectance-mode MW-PPG signals from wrist-wearable devices accurately estimate total hemoglobin concentration compared to reference devices in volunteers?
Machine-learning analysis of reflectance-mode MW-PPG signals from wrist-wearable devices can estimate total hemoglobin concentration with accuracy comparable to clinical noninvasive devices.
Answers are grounded in this paper and cite their sources.
Wrist PPG with ML may enable noninvasive hemoglobin monitoring; leaves open prospective validation before clinical use.
This paper describes the application of multiwavelength photoplethysmography (MW-PPG) in reflectance mode for noninvasive measurements of total hemoglobin concentration. Consumer wrist-wearable devices (smartwatches, wristbands) typically contain a photoplethysmography sensor operating in reflectance mode, as opposed to clinical pulse-oximetry sensors operating in transmittance mode. We assume that the generally accepted approach to the analysis of the transmittance-mode MW-PPG signal, based on the direct implementation of the Beer-Lambert law, is not applicable to the analysis of reflectance-mode MW-PPG signals. We propose that the shape of the MW-PPG signal carries information regarding the distribution of optical paths at different wavelengths within the tissue, which is crucial for the correct estimation of the absorption of light in the probe volume. We have developed and tested several machine-learning algorithms to analyze the reflectance-mode MW-PPG signal and to estimate the total hemoglobin concentration based on this analysis. To train and validate the algorithms, we collected a dataset of 840 MW-PPG signals measured from 170 volunteers by using a wrist-wearable PPG sensor. Three reference devices were used to label the data and test the performance of the developed algorithms: an invasive laboratory blood test, minimally invasive HemoCue Hb 201+, and noninvasive Masimo Radical-7 transmittance mode pulse-oximeter. The best performance achieved with the developed algorithm, mean absolute error MAE ≈ 12.6±1.7 g/L and correlation coefficient R ≈ 0.66±0.09, is comparable with the performance of the clinical noninvasive device.
No takes yet. Share an insight, caveat, or question.
Лычагов et al. (2023) reported an observational. Reflectance-mode multiwavelength photoplethysmography (MW-PPG) with machine-learning analysis vs. Invasive laboratory blood test, HemoCue Hb 201+, and Masimo Radical-7 pulse-oximeter was evaluated on Total hemoglobin concentration estimation accuracy (MAE 12.6 g/L, R 0.66). Machine-learning analysis of reflectance-mode wrist photoplethysmography estimated total hemoglobin concentration with a mean absolute error of 12.6±1.7 g/L and correlation coefficient of 0.66±0.09.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: