Title: A Meta-Analysis of the Impact of Skin Type and Gender on Non-contact Photoplethysmography Measurements
It is well established that many datasets used for computer vision tasks are not representative and may be biased. The result of this is that evaluation metrics may not reflect real-world performance and might expose some groups (often minorities) to greater risks than others. Imaging photoplethysmography is a set of techniques that enables noncontact measurement of vital signs using imaging devices. While these methods hold great promise for low-cost and scalable physiological monitoring, it is important that performance is characterized accurately over diverse populations. We perform a meta-analysis across three datasets, including 73 people and over 400 videos featuring a broad range of skin types to study how skin types and gender affect the measurements. While heart rate measurement can be performed on all skin types under certain conditions, we find that average performance drops significantly for the darkest skin type. We also observe a slight drop in the performance for females. We compare supervised and unsupervised learning algorithms and find that skin type does not impact all methods equally. The imaging photoplethysmography community should devote greater efforts to addressing these disparities and collecting representative datasets. more »« less
Wong, Anthony Y.; Geddes, Jeffrey A.; Tai, Amos P.; Silva, Sam J.(
, Atmospheric Chemistry and Physics)
Abstract. Dry deposition is a major sink of tropospheric ozone.Increasing evidence has shown that ozone dry deposition actively linksmeteorology and hydrology with ozone air quality. However, there is littlesystematic investigation on the performance of different ozone drydeposition parameterizations at the global scale and how parameterizationchoice can impact surface ozone simulations. Here, we present the results ofthe first global, multidecadal modelling and evaluation of ozone drydeposition velocity (vd) using multiple ozone dry depositionparameterizations. We model ozone dry deposition velocities over 1982–2011using four ozone dry deposition parameterizations that are representative ofcurrent approaches in global ozone dry deposition modelling. We useconsistent assimilated meteorology, land cover, and satellite-derived leafarea index (LAI) across all four, such that the differences in simulatedvd are entirely due to differences in deposition model structures orassumptions about how land types are treated in each. In addition, we usethe surface ozone sensitivity to vd predicted by a chemical transportmodel to estimate the impact of mean and variability of ozone dry depositionvelocity on surface ozone. Our estimated vd values from four differentparameterizations are evaluated against field observations, and whileperformance varies considerably by land cover types, our results suggestthat none of the parameterizations are universally better than the others.Discrepancy in simulated mean vd among the parameterizations isestimated to cause 2 to 5 ppbv of discrepancy in surface ozone in theNorthern Hemisphere (NH) and up to 8 ppbv in tropical rainforests in July,and up to 8 ppbv in tropical rainforests and seasonally dry tropical forestsin Indochina in December. Parameterization-specific biases based onindividual land cover type and hydroclimate are found to be the two maindrivers of such discrepancies. We find statistically significant trends inthe multiannual time series of simulated July daytime vd in allparameterizations, driven by warming and drying (southern Amazonia, southernAfrican savannah, and Mongolia) or greening (high latitudes). The trend inJuly daytime vd is estimated to be 1 % yr−1 and leadsto up to 3 ppbv of surface ozone changes over 1982–2011. The interannual coefficient ofvariation (CV) of July daytime mean vd in NH is found to be5 %–15 %, with spatial distribution that varies with the dry depositionparameterization. Our sensitivity simulations suggest this can contributebetween 0.5 to 2 ppbv to interannual variability (IAV) in surface ozone, butall models tend to underestimate interannual CV when compared to long-termozone flux observations. We also find that IAV in some dry depositionparameterizations is more sensitive to LAI, while in others it is more sensitiveto climate. Comparisons with other published estimates of the IAV ofbackground ozone confirm that ozone dry deposition can be an important partof natural surface ozone variability. Our results demonstrate the importanceof ozone dry deposition parameterization choice on surface ozone modellingand the impact of IAV of vd on surface ozone, thus making a strong casefor further measurement, evaluation, and model–data integration of ozone drydeposition on different spatiotemporal scales.
Cannon, Taylor M.; Bouma, Brett E.; Uribe-Patarroyo, Néstor(
, Biomedical Optics Express)
Structural optical coherence tomography (OCT) images of tissue stand to benefit from greater functionalization and quantitative interpretation. The OCT attenuation coefficientµ, an analogue of the imaged sample’s scattering coefficient, offers potential functional contrast based on the relationship ofµto sub-resolution physical properties of the sample. Attenuation coefficients are computed either by fitting a representativeµover several depth-wise pixels of a sample’s intensity decay, or by using previously-developed depth-resolved attenuation algorithms by Girardet al.[Invest. Ophthalmol. Vis. Sci.52,7738(2011).10.1167/iovs.10-6925] and Vermeeret al.[Biomed. Opt. Express5,322(2014).10.1364/BOE.5.000322]. However, the former method sacrifices axial information in the tomogram, while the latter relies on the stringent assumption that the sample’s backscattering fraction, another optical property, does not vary along depth. This assumption may be violated by layered tissues commonly observed in clinical imaging applications. Our approach preserves the full depth resolution of the attenuation map but removes its dependence on backscattering fraction by performing signal analysis inside individual discrete layers over which the scattering properties (e.g., attenuation and backscattering fraction) vary minimally. Although this approach necessitates the detection of these layers, it removes the constant-backscattering-fraction assumption that has constrained quantitative attenuation coefficient analysis in the past, and additionally yields a layer-resolved backscattering fraction, providing complementary scattering information to the attenuation coefficient. We validate our approach using automated layer detection in layered phantoms, for which the measured optical properties were in good agreement with theoretical values calculated with Mie theory, and show preliminary results in tissue alongside corresponding histological analysis. Together, accurate backscattering fraction and attenuation coefficient measurements enable the estimation of both particle density and size, which is not possible from attenuation measurements alone. We hope that this improvement to depth-resolved attenuation coefficient measurement, augmented by a layer-resolved backscattering fraction, will increase the diagnostic power of quantitative OCT imaging.
Rodriguez, Andres J.; Vasudevan, Sandhya; Farahmand, Masoud; Weininger, Sandy; Vogt, William C.; Scully, Christopher G.; Ramella-Roman, Jessica; Pfefer, T. Joshua(
, Biomedical Optics Express)
Pulse oximetry represents a ubiquitous clinical application of optics in modern medicine. Recent studies have raised concerns regarding the potential impact of confounders, such as variable skin pigmentation and perfusion, on blood oxygen saturation measurement accuracy in pulse oximeters. Tissue-mimicking phantom testing offers a low-cost, well-controlled solution for characterizing device performance and studying potential error sources, which may thus reduce the need for costly in vivo trials. The purpose of this study was to develop realistic phantom-based test methods for pulse oximetry. Material optical and mechanical properties were reviewed, selected, and tuned for optimal biological relevance, e.g., oxygenated tissue absorption and scattering, strength, elasticity, hardness, and other parameters representing the human finger’s geometry and composition, such as blood vessel size and distribution, and perfusion. Relevant anatomical and physiological properties are summarized and implemented toward the creation of a preliminary finger phantom. To create a preliminary finger phantom, we synthesized a high-compliance silicone matrix with scatterers for embedding flexible tubing and investigated the addition of these scatterers to novel 3D printing resins for optical property control without altering mechanical stability, streamlining the production of phantoms with biologically relevant characteristics. Phantom utility was demonstrated by applying dynamic, pressure waveforms to produce tube volume change and resultant photoplethysmography (PPG) signals. 3D printed phantoms achieved more biologically relevant conditions compared to molded phantoms. These preliminary results indicate that the phantoms show strong potential to be developed into tools for evaluating pulse oximetry performance. Gaps, recommendations, and strategies are presented for continued phantom development.
Nadel, Brian B; Oliva, Meritxell; Shou, Benjamin L; Mitchell, Keith; Ma, Feiyang; Montoya, Dennis J; Mouton, Alice; Kim-Hellmuth, Sarah; Stranger, Barbara E; Pellegrini, Matteo; et al(
, Briefings in Bioinformatics)
Abstract Estimating cell type composition of blood and tissue samples is a biological challenge relevant in both laboratory studies and clinical care. In recent years, a number of computational tools have been developed to estimate cell type abundance using gene expression data. Although these tools use a variety of approaches, they all leverage expression profiles from purified cell types to evaluate the cell type composition within samples. In this study, we compare 12 cell type quantification tools and evaluate their performance while using each of 10 separate reference profiles. Specifically, we have run each tool on over 4000 samples with known cell type proportions, spanning both immune and stromal cell types. A total of 12 of these represent in vitro synthetic mixtures and 300 represent in silico synthetic mixtures prepared using single-cell data. A final 3728 clinical samples have been collected from the Framingham cohort, for which cell populations have been quantified using electrical impedance cell counting. When tools are applied to the Framingham dataset, the tool Estimating the Proportions of Immune and Cancer cells (EPIC) produces the highest correlation, whereas Gene Expression Deconvolution Interactive Tool (GEDIT) produces the lowest error. The best tool for other datasets is varied, but CIBERSORT and GEDIT most consistently produce accurate results. We find that optimal reference depends on the tool used, and report suggested references to be used with each tool. Most tools return results within minutes, but on large datasets runtimes for CIBERSORT can exceed hours or even days. We conclude that deconvolution methods are capable of returning high-quality results, but that proper reference selection is critical.
Camera-based physiological measurement enables vital signs to be captured unobtrusively without contact with the body. Remote, or imaging, photoplethysmography involves recovering peripheral blood flow from subtle variations in video pixel intensities. While the pulse signal might be easy to obtain from high quality uncompressed videos, the signal-to-noise ratio drops dramatically with video bitrate. Uncompressed videos incur large file storage and data transfer costs, making analysis, manipulation and sharing challenging. To help address these challenges, we use compression specific supervised models to mitigate the effect of temporal video compression on heart rate estimates. We perform a systematic evaluation of the performance of state-of-the-art algorithms across different levels, and formats, of compression. We demonstrate that networks trained on compressed videos consistently outperform other benchmark methods, both on stationary videos and videos with significant rigid head motions. By training on videos with the same, or higher compression factor than test videos, we achieve improvements in signal-to-noise ratio (SNR) of up to 3 dB and mean absolute error (MAE) of up to 6 beats per minute (BPM).
Nowara, Ewa M., McDuff, Daniel, and Veeraraghavan, Ashok. A Meta-Analysis of the Impact of Skin Type and Gender on Non-contact Photoplethysmography Measurements. Retrieved from https://par.nsf.gov/biblio/10217885. A Meta-Analysis of the Impact of Skin Type and Gender on Non-contact Photoplethysmography Measurements . Web. doi:10.1109/CVPRW50498.2020.00150.
Nowara, Ewa M., McDuff, Daniel, & Veeraraghavan, Ashok. A Meta-Analysis of the Impact of Skin Type and Gender on Non-contact Photoplethysmography Measurements. A Meta-Analysis of the Impact of Skin Type and Gender on Non-contact Photoplethysmography Measurements, (). Retrieved from https://par.nsf.gov/biblio/10217885. https://doi.org/10.1109/CVPRW50498.2020.00150
Nowara, Ewa M., McDuff, Daniel, and Veeraraghavan, Ashok.
"A Meta-Analysis of the Impact of Skin Type and Gender on Non-contact Photoplethysmography Measurements". A Meta-Analysis of the Impact of Skin Type and Gender on Non-contact Photoplethysmography Measurements (). Country unknown/Code not available. https://doi.org/10.1109/CVPRW50498.2020.00150.https://par.nsf.gov/biblio/10217885.
@article{osti_10217885,
place = {Country unknown/Code not available},
title = {A Meta-Analysis of the Impact of Skin Type and Gender on Non-contact Photoplethysmography Measurements},
url = {https://par.nsf.gov/biblio/10217885},
DOI = {10.1109/CVPRW50498.2020.00150},
abstractNote = {It is well established that many datasets used for computer vision tasks are not representative and may be biased. The result of this is that evaluation metrics may not reflect real-world performance and might expose some groups (often minorities) to greater risks than others. Imaging photoplethysmography is a set of techniques that enables noncontact measurement of vital signs using imaging devices. While these methods hold great promise for low-cost and scalable physiological monitoring, it is important that performance is characterized accurately over diverse populations. We perform a meta-analysis across three datasets, including 73 people and over 400 videos featuring a broad range of skin types to study how skin types and gender affect the measurements. While heart rate measurement can be performed on all skin types under certain conditions, we find that average performance drops significantly for the darkest skin type. We also observe a slight drop in the performance for females. We compare supervised and unsupervised learning algorithms and find that skin type does not impact all methods equally. The imaging photoplethysmography community should devote greater efforts to addressing these disparities and collecting representative datasets.},
journal = {A Meta-Analysis of the Impact of Skin Type and Gender on Non-contact Photoplethysmography Measurements},
author = {Nowara, Ewa M. and McDuff, Daniel and Veeraraghavan, Ashok},
editor = {null}
}
Warning: Leaving National Science Foundation Website
You are now leaving the National Science Foundation website to go to a non-government website.
Website:
NSF takes no responsibility for and exercises no control over the views expressed or the accuracy of
the information contained on this site. Also be aware that NSF's privacy policy does not apply to this site.
