J. Biophoton. 2, No. 5, 277–287 (2009) / DOI 10.1002/jbio.200910019
Journal of
BIOPHOTONICS
REVIEW ARTICLE
Photonics-based In Vivo total hemoglobin monitoring
and clinical relevance
John McMurdy1, Gregory Jay1; 2, Selim Suner1; 2, and Gregory Crawford *; 1; 3
1
2
3
Division of Engineering, Brown University, Providence, RI 02912, USA
Department of Emergency Medicine, Rhode Island Hospital, Providence, RI 02903, USA
College of Science and Department of Physics, University of Notre Dame, Notre Dame, Indiana 46556, USA
Received 28 October 2008, revised 10 February 2009, accepted 6 March 2009
Published online 22 April 2009
Key words: Anemia, hemoglobin, spectroscopy, tissue optics, noninvasive
PACS: 81.15.M, 82.80.d, 87.85.Ox, 87.85.Pq
Anemia is a serious disorder which, as a result of antiquated invasive blood testing, is undiagnosed in millions
of people in the U.S. As a result of the clinical need,
many technological solutions have been proposed to
measure total blood hemoglobin, and thus diagnose anemia, noninvasively. Because hemoglobin is the strongest
chromophore in tissue, spectroscopic methods have been
the most prevalently investigated. Difficulties in extracting a quantitative estimation of hemoglobin based on
tissue absorption include variability in the absorption
spectra of hemoglobin derivatives, interference from
other tissue chromophores, and interpatient physiological variations affecting the effective optical pathlength of
light propagating in tissue. In spite of these challenges,
studies with a high degree of correlation between in vitro and in vivo measured total hemoglobin have been
disclosed using variants of transmission and diffuse reflection spectroscopy in assorted physiological locations.
A review of these technologies and the relevant advantages/disadvantages are presented here.
Reflectance image of blood vessels in the sublingual mucosa from a healthy (top) and anemia (bottom) patient collected using the orthogonal polarization spectral imaging
technique to enhance vessel contrast. Adapted from [7].
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
* Corresponding author: e-mail: gregory_crawford@nd.edu
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Journal of
BIOPHOTONICS
278
1. Introduction
Anemia is clinically defined as a significant reduction (two standard deviations) in either the number
of red blood cells (RBCs) per unit volume of whole
blood, the amount of packed RBCs per unit volume
of whole blood, or most generally the amount of the
oxygen/CO2 transporting protein hemoglobin (Hgb)
from the clinically determined ‘normal’ value [8]. It
is vital to screen for anemia as it has been reported
to affect 2 billion people worldwide, or roughly onethird of the worlds population [9]. Anemia is a complication of dietary deficiencies including iron deficiency, vitamin B12 deficiency, and folic acid deficiency, but also appears concurrent with more
serious disorders including cancer, HIV/AIDS, leukemia, chronic kidney disease, and rheumatoid arthritis. The appearance of anemia with numerous
disorders makes Hgb concentration monitoring a vital parameter in assessing overall public health.
Invasive determination of blood hemoglobin levels through the complete blood count (CBC) remains the most clinically embraced method of anemia detection. Parameters of the CBC are obtained
using biochemical assays, flow cytometry, and analysis of blood smears, contemporarily in a fully integrated workstation. Although remaining invaluable
as a test for many specific anemias, CBCs are time
consuming, costly, and increase risks of phlebotimist
exposure to blood-borne pathogens. If the full CBC
results are not necessary, blood can be centrifuged in
a small volume device at the bedside providing a hematocrit (Hct, volume fraction of RBC’s and roughly
correlated with Hgb) determination; however, this
remains an invasive technique and creates cross-contamination possibilities for neighboring patients and
physicians.
In the contemporary climate of non-invasive diagnostics, the CBC and finger prick methods remain
antiquated. While it is not anticipated that non-invasive techniques in their infancy will provide as much
diagnostic information as the CBC, screening technologies will alert physicians to a patients Hgb level
and encourage further investigation. Repeated anemia screening is necessary high-risk demographics
including pregnant women, children, and the elderly,
all drawing great benefit from rapid non-invasive anemia detection. This need propelled studies as early
as the 80’s on conductive and early optical techniques of total hemoglobin/hematocrit monitoring
[1, 6]. More recently, the necessity of enhanced techniques to rapidly screen for anemia has resulted in
investigations on numerous non-invasive methods,
the majority of which are based on interrogation of
light-tissue interactions. Presented here is a brief discussion of the clinical decision making altered by the
presence of such advice, the fundamental tissue spec-
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
J. McMurdy et al.: In vivo hemoglobin determination
troscopy challenges to overcome in extracting Hgb
concentration, and a survey of the most recent
photonics-based methods of detection reported in
the literature.
2. Clinical decision making
There are numerous clinical circumstances where rapid determination of Hgb is vital and other situations where immediate knowledge of Hgb can be
used to enhance patient care. With the development
of non-invasive, rapid, low cost, painless devices
which could be easily used with little training, Hgb
can be added to the list of variables collected from
each patient during every health care encounter (vital signs) such as blood pressure, heart rate, temperature, respiratory rate, blood oxygen saturation
(SpO2 ) and blood glucose concentration. Any entry
point to the health care system such as emergency
departments, clinics and doctors offices universally
measure vital signs. In the future, insertion of technological advances in engineering, materials science,
photonics and chemistry to the bedside will expand
this list.
Patients who arrive to the emergency department, of which there were 119.2 million during 2006
in the United States, require proper triage to determine which group must be seen immediately [10].
The median time for emergency department visits is
2.6 hours with only 21.9% of patients seen within
15 minutes of arrival. Objective data such as vital
signs, available to the clinician, make the rapid triage
decision more accurate. Adding Hgb to vital signs
obtained at the point of triage will enhance the
triage process and improve patient care by immediately identifying patients with severe anemia and reducing their wait times. In most instances, new technologies related to vital sign monitoring are rapidly
adopted by the emergency medical services community and integrated into ambulance services, indeed
the use of electrocardiography and pulse oximetry
were rapidly adopted by EMS shortly after being implemented and most recently carboxyhemoglobin
monitoring has been incorporated into EMS practice
almost simultaneously with hospital based health
care systems. The use of these technologies by EMS
allows objective data to be collected in the ambulance even before the patient arrives in the emergency department. We envision light, non-invasive,
inexpensive devices that measure Hgb will follow
this trend and make it into the pre-hospital setting.
Anemia is most prevalent in developing countries
where nutritional resources and access to health are
lacking. Also, these regions of the world have parasitic and other diseases, which contribute to anemia.
Anemia leads to higher infant mortality, loss of man-
www.biophotonics-journal.org
REVIEW
ARTICLE
J. Biophoton. 2, No. 5 (2009)
power and productivity in the work force and exacerbates co-morbid disorders such as heart disease.
These are precisely the communities where wide
spread surveillance could be used to identify anemia
and institute treatment measures. However, current
screening and detection methods rely on an established healthcare infrastructure, which is often lacking in these regions of the world. An inexpensive and
point of care device to measure Hgb could be deployed in this context for surveillance. Non-invasive
Hgb measurements can also be used for continuously
monitoring patients with frequent repeated testing
since the Hgb in the blood changes slowly (on the
order of hours, even with massive blood loss). This
application can be particularly useful for those patients with on-going hemorrhage, those receiving
blood transfusion or hemodialysis and patients undergoing major surgical procedures.
It is difficult to assess what a clinically acceptable
level of accuracy in monitoring total hemoglobin
concentration may be. This is directly tied to the intended use of the device and the individual making
a decision on the patients care. For example, in using
the device as a continuous monitor of total hemoglobin, the required error levels may be significantly
less than 1.0 g/dL in order to observe trends. Alternatively, in using the device as a rapid screening
tool, an error level of 1.0 g/dL may be tolerable as
subsequent testing will be required. The situation
and how the information will change a patients
course of treatment will ultimately determine the acceptance of these performance levels.
279
A ¼ " c l, where " is the molar extinction coefficient of the analyte, c is the concentration of the
analyte, l is the optical pathlength integrating the refractive index. While this is a straightforward calculation in an in vitro controlled environment, in an in
vivo tissue model it becomes significantly more complex. In the case of a transmission measurement system, cellular structures such as nuclei and cell membranes contribute a significant amount of optical
scatter to the system having the effect of artificially
increasing the optical pathlength. As the scattering
properties are wavelength dependent, changes in
scatter not only have the effect of changes overall
intensity, but distort spectral signatures through wavelength dependent pathlength variation.
Additionally, the structure itself may vary in
thickness (e.g. changing finger/earlobe thickness
from person to person) while the volume of blood
experienced during propagation can also vary as a
result of both physiological variability and from pulsatilechanges in vessel diameter. Diffuse reflectance
measurements suffer from these same sources of error, explicitly changing blood volumes and a combination of scattering and layer thickness variations affecting the average penetration depth of light into
the tissue and distorting spectral signatures. What results are equations for transmission and reflected intensity calculated from steady state diffusion theory
with a dependence on numerous parameters as
shown in Eq. (1).
Rd ðlÞ ¼ e
B
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3ð1þms ðlÞ ð1gÞ ma ðlÞÞ
ð1Þ
There exists fundamental technological challenges in
determining Hgb concentration quantitatively using
an optical or photonics based solution. In general,
these difficulties arise from inter-patient variation in
physiological characteristics causing significant deviation in experimental conditions from one patient
to the next. Three of the most significant concerns in
quantitatively measuring Hgb are fluctuations in the
concentrations of melanin and other chromophores,
the presence of several hemoglobin derivatives, and
changes in optical path length used in a Beer-Lambert law calculation.
Here, I1 and I0 are the output and input intensities,
respectively, ma is the absorption coefficient where
ma " c, ms is the scattering coefficient describing
the concentration of scattering particles and their
scattering efficiency (which may or may not be the
same as absorption particles), g is the scattering anisotropy describing the directionality of scatter, and
B is some tissue dependent constant. Explicitly modeling all of these parameters for each patient is not
feasible, thus alternative methods of reconciling this
source of error have been explored, as are discussed
here. One such technique may be using the absorption spectra of water or another biomarker to calibrate for this pathlength variability; however, this
has yet to be successfully implemented in conjunction with hemoglobin monitoring.
3.1. Optical pathlength
3.2. Hemoglobin variants
Determination of the unknown concentration of an
analyte of interest through diffuse reflectance or
transmissive spectroscopy methods is based upon the
Beer-Lambert law, which states the absorption
Although differences in the absorption spectra
between hemoglobin species are exploited to determine relative amounts of each in saturation measurements, they acts as an interference to determining
3. Technology considerations
www.biophotonics-journal.org
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Journal of
BIOPHOTONICS
280
Figure 1 Absorption spectra in the NIR regime of oxy-,
deoxy-, met-, and carboxyhemoglobin. Adapted from [11].
the summation of these species, termed total hemoglobin (Hgb) concentration. The primary derivatives of hemoglobin that may be observed in significant concentration levels in many disorders are
methemoglobin (MetHb) and carboxyhemoglobin
(HbCO), both of which have differing absorption
spectra from normal oxyhemoglobin (HbO2 ) and
deoxyhemoglobin (Hb). As is the case in normal
hemoglobin, these derivates vary significantly in
the NIR spectral regime (as pictures in Figure 1)
and there exist no isosbestic points between all
four. As a result, extraction of total hemoglobin
(Hgb ¼ HbO2 þ Hb þ HbCO þ MetHb) becomes
more complex requiring many wavelength bands to
discern these species.
J. McMurdy et al.: In vivo hemoglobin determination
amount of oxyhemoglobin to oxy- plus deoxyhemoglobin has the effect of normalizing the pathlength
term and all other constant terms out of the equations, making determination of the saturation capable in a highly variable population. Determining
the actual concentration of one of these species, or
the summation for total hemoglobin, unfortunately
does not have the same benefit ofpatient independence, and thus in this case the same predictive
model cannot be utilized. However, recent efforts
have been made to vary pulse oximetry optics and
modify signal processing to create a CO-oximeter
capable of using comparable pulsatile signals to determine methemoglobin [12], carboxyhemoglobin
[13] concentration, and total hemoglobin concentration [14]. This technology along with other technologies to resolve the pathlength fluctuation issue are
discussed later in Section 4.
In spite of the challenges, many groups have investigated alternative technologies to try and achieve
the success of the pulse oximeter for Hgb. This approaches are broadly categorized here into two
classes, NIR transmission spectroscopy systems and
visible/NIR diffuse reflectance spectroscopy systems.
A summary of these devices is presented in the following two sections. It is important to note that the
accuracy and correlation values are given as a gross
reference point and not an absolute indicator of the
device potential. These studies were all carried out
under different clinical conditions with different patient populations, different invasive methods used as
benchmarks, and different comparison techniques.
3.3. Melanin and other interference analytes
4. Transmission spectroscopy technologies
Complicating interpatient changes in optical pathlength are fluctuations of other chromophores unrelated to Hgb. Primary chromophores in the visible
spectral regime, in addition to total Hgb and its derivates, include melanin, various flavins, cytochrome-coxidase, carotenoids and extravasated hemoglobin
and its derivatives induced through hemorrhage or
contusion. Absorption at longer wavelength NIR radiation also includes contributions from myoglobin,
other cytochromes, and water (at wavelengths longer
than 1 mm). Although the extinction coefficient of
most of these species is small compared to that of
Hgb, the net effect of the sum of all these factors
can vary significantly from patient to patient and
from tissue to tissue, creating a significant source of
error to extracting pure Hgb quantitative information.
Interestingly, pulse oximetry overcomes many of
these interpatient variations to determine systolic
pulse hemoglobin saturation by performing calculation based on the change in absorption during systole and diastole periods. Looking at the relative
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
The primary physiological location used for transmission spectroscopy quantification of Hgb is the fingertip [14–23], enabling many similar probe configurations utilized in pulse oximetry to be exploited.
A brief discussion of the state of art in these devices
and their ability to overcome the challenges discusses in Section 3 is presented here.
One method of directly monitoring changes in
optical pathlength through the fingertip is reported
by Aldrich et. al. [21]. NIR transmittance through
the fingertip at a single wavelength (905 nm, non-isosbestic for oxy- and deoxyhemoglobin) is used in
conjunction with acoustic transducers to kinetically
quantify both the bulk thickness of the tissue segment and small variations in this thickness from
blood vessel dilation during systole. Sonomicrometers are positioned at opposite sides of the finger to
estimate the effective optical pathlength and subsequent pulsatile pathlength variation, as is shown in
Figure 2(a). Transmitted intensity of the 905 nm NIR
LED through the finger at this same position is then
www.biophotonics-journal.org
REVIEW
ARTICLE
J. Biophoton. 2, No. 5 (2009)
Figure 2 Setup of transcutaneous illumination coupled
with path length modulation detection using sonomicrometer transducers (a) and the detected variation in light intensity corresponding to the sonomicrometer detected variation in optical pathlength (b). Adapted from [21].
normalized by this acoustically determined optical
pathlength and correlated to an in vitro Hgb determination. Figure 2(b) shows this pathlength and
transmission variation during systolic pulsation. This
group reports a correlation r ¼ 0.84 (n ¼ 24) for
Hgb, sensitivity and specificity of 94% and 78% for
detecting anemia respectively, and mean predictive
error of 1.1 g/dL for Hgb. While the acoustic method
of pathlength monitoring can account for interpatient fingertip thickness changes through signal intensity normalization, sources of error remain from
varying blood volume fractions, scattering properties,
and interfering analytes which are not rejected in
the processing method currently described. It remains unclear in this case (1) whether the difference
in transmitted intensity between the minimum and
maximum sonomicrometer determined pathlength
can be used to normalize for tissue variation and
correlate to Hgb and (2) how the use of sonomicrometers might be suited ina true clinical application.
Jeon et. al. disclosed an alternative method of
pathlength correction in fingertip transmissive Hgb
measurement using the pulse varying optical signal
themselves [20]. On transcutaneous illumination of
the finger with visible/NIR diodes, as light diffuses
through the fingertip, a fraction interacts with the arterial vessels while another fraction interacts only
with veins and soft tissue in the fingertip, as is shown
in Figure 3. As a result, photons reaching the photodetectorafter passing through an arterial vessel are
subject to pulsatile modulation in intensity from
changing vessel diameter (and subsquent interaction
www.biophotonics-journal.org
281
length with whole blood) while photons bypassing
the arterial vessel remain constant during pulsatile
waveforms. The ratio of pulsatile varying signals to
constant signals is then used to minimize effects
from pathlength variation and interference from
other analytes. This group initially reported correlation r ¼ 0.80 (n ¼ 129) and standard deviation of error (standard error) of 1.14 g/dL for Hgb relative to
in vitro determined levels using the aforementioned
ratio as a correlation variable [20] and has since estimated potential clinical performance to a correlation
of r ¼ 0.87 and standard error of 0.81 g/dL with a
more optimized finger probe [24]. This degree of
performance is achieved using a five diode array
with multiple wavelength ratios found to most effectively reject additional errors from saturation state
and wavelength dependent tissue scattering changes.
Additionally, other earlier studies have shown similar signal processing techniques to extract Hgb concentration using pulsatile variation in pathlength in
transmission through the earlobe with correlation
coefficient r ¼ 0.79 (n ¼ 14) [25].
Although the change in vessel diameter and thus
blood volume during systole provides a method to
determine Hgb in the presence of other varying factors, the change in vessel diameter is often only a
few percent, making this signal modulation relatively
weak. An alternative method is to create an artificial
variation in local Hgb concentration and thus transmitted intensity by artificially controlling blood flow.
Enhancement though artificial blood volume fraction
modulation has been explored by several groups.
Gravenstein et. al. was one of the earliest to discuss this type of measurement when disclosing a canine animal model using volumetric constriction of
blood flow to measure Hgb noninvasively [23]. An
810 nm diode laser was transilluminated through a
canine tongue while a micrometer was used to restrict blood flow through applied pressure. Diluting
Hgb concentration using an intravenous NaCl solution, variations in transmitted signal were correlated
to applied pressure in each canineleading to a corre-
Figure 3 Schematic of light diffusing through the fingertip
artery showing a fractional portion propagating through a
pulsatile varying vessel and fraction remaining stable with
pulse. Adapted from [20].
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Journal of
BIOPHOTONICS
282
lation range of r ¼ 0:87 0:97. Although this technique showed efficacy in dynamically monitoring Hgb,
discrepancy between the correlation of NIR transmission with applied pressure precluded an assessment of the ability of this technique to measure interpatient Hgb concentration noninvasively [23].
More contemporarily, Rendell et. al. has explored
modulating NIR transmission through the fingertip
using inserts that volumetrically constrict and release
blood flow [19]. The modulation in transmitted intensity with different inserts is then correlated to Hgb
using a linear regression method. A regression algorithm applied to the artificially modulated signal intensities achieves an optimal correlation r ¼ 0.862
(n ¼ 121) and mean error of prediction of 1.28 g/dL
for Hgb compared to invasively determined levels. A
second group has explored a similar method they term
occlusion spectroscopy which utilizes combined fingertip transmission specroscopy and blood flow modulation [26, 27, 15]. Transmitted intensity is varied by
occluding blood flow using a restrictive finger cuff (similar to a blood pressure cuff) accelerating ischemia
driven red blood cell aggregation. The occluding finger cuff is later released allowing red blood cells to
disaggregate while the changein optical transmission
at multiple NIR/visible wavelengths is monitored [28].
This enhancement in contrast using the difference of
these two signals enables a more accurate parameter
correlation to Hgb. Figure 4 shows an in vitro demonstration of the change in blood transmission, attributed to the variation in scattering properties, observed
during pulsations of blood in a capillary tube. The
most recently reported clinical performance of this
technology shows a correlation coefficeint r ¼ 0.9
(n ¼ 309) and standard error values of 3.3% (for hematocrit), or 1.0 g/dL (using the relationship Hgb(g/
dL) ¼ 0.34 Hct(%)) [15]. A third variant of this class
of technology also uses a finger cuff to modulate the
volume of blood flow into the finger and measure
transcutaneous optical absorptions to infer Hct, having shown a correlation r ¼ 0:78 ðn ¼ 121Þ for Hct in
the most recent published results [16].
Figure 4 In Vitro demonstration of the change in optical
scattering and thus transmission during pulsatile action attributed to variation in red blood cell aggregates. Adapted
from [28].
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
J. McMurdy et al.: In vivo hemoglobin determination
Alternatively, implementing an imaging device in
a pulse oximetry probe configuration allows for a direct determination of Hgb by monitoring the transmitted intensity and imaging the blood vessels
probed in vivo. Kanashima et. al. has recently reviewed the performance of a NIR imaging based
noninvasive Hgb monitor in patients with and without hematological disorders [17, 29, 22]. Photometric
absorbance measurements from the vascular portions of the image using multiple discrete NIR bands
are coupled with an estimation of the blood vessel
optical pathlength. This pathlength is estimated
based on in vivo imaged vascular diameter, and assuming circular vessel cross sections. Intensity measurements can then be normalized for changing optical pathlength, and Hgb subsequently correlated to
this parameter, again using regression analysis. This
group discloses sensitivity and specificity of 78.3%
and 69% for clinically defined anemia, respectively,
with correlation r ¼ 0.53 for anemic patients
(n ¼ 174), correlation r ¼ 0.34 (n ¼ 135) for patients
with normal Hgb levels, and an overall correlation
r ¼ 0:591 ðn ¼ 309Þ for all measurements [17]. It is
unclear why this recent study has shown this device
tobe so much more inaccurate than an earlier report
disclosing a correlation r ¼ 0.896 (n ¼ 91) [22].
While this technique has the benefit of correction
for blood vessel dilation with pulse, a significant
source of error remains due to other chromophores
influencing transmitted intensity independent of Hgb
and vessel diameter. This is likely why the authors
report a high precision in single patient Hgb tracking, but observe a significant drop in accuracy in determining interpatient Hgb concentration from a single reading.
5. Reflectance spectroscopy
and imaging technologies
Various approaches using diffuse reflectance have
also been examined. Reflectance spectroscopy and
imaging can be advantageous as deep transmission
through highly scattering turbid tissue is not necessary, but rather properties of blood vessels proximal
to the surface can be probed. Unfortunately, issues
still arise from other absorption events affecting reflected signal. Also, the pathlength variation problem persists, manifesting as a variability in photon
penetration depth that is dependent on tissue layers
absorption, reflectance, and scattering properties,
and can result in nonlinear intensity responses with
linear variation in Hgb concentration. Reflectance
based methods mitigating these challenges can be
grouped as either reflectance and spectral reflectance imaging based modes or single point detection
reflection spectroscopy modes.
www.biophotonics-journal.org
REVIEW
ARTICLE
J. Biophoton. 2, No. 5 (2009)
283
5.1. Imaging based modalities
In the case of highly superficial blood vessels, as is
the case in many vessels of mucosal surfaces (retina,
sublingual mucosa, tongue, conjunctiva), hemoglobin
concentration can be estimated through reflectance
imaging with multiple wavelength bands and subsequent image processing analysis. Mucosal blood vessels are the most desirable access point to estimate
blood component parameters due to the lack of thick
overlaying tissues, and subsequent minimization of
other interfering analytes not related to whole blood.
One of the most interesting imaging based methods of in vivo Hgb determination is orthogonal polarization spectral (OPS) image analysis of the vascular network in the sublingual mucosal surface [7,
30–32]. Hemoglobin is determined based upon a
computational estimation of vascular network density from collected images and the intensity of reflectance signal from the vessel area. Images collected
using OPS imaging of the microvascular network for
anemic and healthy patients are shown in Figure
5(a–d). A polarized light source in a spectral region
of high hemoglobin absorbance (550 nm) is used
to illuminate the vasculature while a polarizer at 90
with respect to incident light polarization is placed at
the detection focal plane. The detection plane polarizer only transmits light that has been depolarized
through scattering after penetrating superficial tissue
layers, enhancing the proportion of photons detected
that interact with microvasculature and rejecting
light reflected through Fresnel reflection at the surface. An image-processing algorithm is then applied
to the OPS image data to estimate Hgb based on the
optical density over a number of individual vessels.
This group has reported a correlation r ¼ 0.93
(n ¼ 71). Interestingly, high resolution OPS images
can also potentially be used to infer morphological
red blood cell characteristics such as mean corpuscu-
Figure 6 Two SD-LCI images of the retina where blood
vessels are denoted by white arrows. Properties such as
the depth and slope of intensity decrease are correlated to
total Hgb. Adapted from [36].
lar volume and abnormal shaped RBC seen in sickle
cell anemia. Complementary standard imaging analysis (not OPS based) has also been applied to both
blood vessels in the retina [33] using multiple discrete NIR/visible bands propagated through the pupil with a correlation r ¼ 0.89 ðn ¼ 24Þ, and much
earlier using blood vessels in the bulbar conjunctiva
through high magnification assessment of capillary
bed absorption [34].
A more complex variant of reflectance imaging,
spectral domain low coherence interferometry (SDLCI) [35], has been used to probe the Hgb dependent absorption and scattering profiles of retinal vessels [36]. SD-LCI is a type of optical coherence tomography, a technique already widely implemented
in retinal vessel imaging to assess various retinal disorders. Penetration depth of photons into a blood
vessel is inversely related to the scattering coefficient, which in turn is correlated to the hematocrit.
This generates a reflectivity depth profile for each
individual blood vessel which, when determined with
an LCI technique, can be used a parameter to correlate with Hct. As most of the light scattered from a
blood vessel is in the forward direction, minimal intensity is projected back into the collection system
and thus long integration/exposure times are required. As a result, a retinal tracking system is also
implemented to maintain fixation on the vessel of interest in the presence of rapid eye movements. Figure 6 shows two SD-LCI images of retinal blood vessels where the vessel absorption notches used to
estimate Hct are indicated. Although this technique
is still in the pilot study stage, it is a creative one that
can be implemented with OCT systems used for
other purposes in an ophthalmology setting. An initial data set of seven volunteers has estimated normal
HCT ranges for both males and female [36].
5.2. Single point detection
Figure 5 Microcirculation images collected using OPS imaging of a healthy patient (a, b) and an anemic patient (c,
d). Reproduced with permission from [7].
www.biophotonics-journal.org
Inferences on hemoglobin concentration can also be
made from light after diffusing through a tissue segment of interest and collecting on a single detector.
These single point measurements are effectively collecting signal over a summation of many diffuse
photon pathlengths. Estimation of hemoglobin con-
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Journal of
BIOPHOTONICS
284
centration is then made based on this signal summation using a regression technique of some sort. In essence, while often more simplistic than transmission
approaches, this indirect method of hemoglobin monitoring the quality of the prediction becomes a function
of the consistency of data collection conditions and
subsequent reliability of the calibration model in a
fluctuating physiological environment. In the scope of
identifying optical properties of the skin, studies have
disclosed the ability to determine contributions to scattering and absorption coefficients attributed to cutaneous hemoglobin [2–5]. The studies discussed here
are limited to those which have disclosed measurement of total hemoglobin concentration concentrations.
In one approach, steady state visible and NIR
diffuse reflectance spectra were collected from the
dorsal side of the arm and correlation to Hgb using
regression methods [37]. Diffuse reflectance spectra
were collected using a 10 nm resolution commercial
diffuse reflectance spectrometer equipped with integrating sphere. A classical least squares linear regression analysis and leave-one-out cross validation
of the absorbance data resulted in correlation
r ¼ 0.8 and standard error of 0.9 g/dL for Hgb prediction. Wavelength regions found to be most correlated to Hgb/Hct were logically similar to those used
optically to assign an erythema index using reflectance spectroscopy. This group interestingly explored
the effects of thermal modulation on tissue diffusion
characteristics, shown to modify absorption and scattering coefficients through changing cutaneous blood
flow [38], and demonstrated improved predictive accuracy by controlling skin temperature correlation
r ¼ 0.8 ðn ¼ 26Þ and standarderror of 0.8 g/dL for
Hgb in patient set with 10 light-skinned patients
tested multiple times (total data points n ¼ 26). This
study also points out the drop in predictive accuracy
with an ethnically diverse patient set, correlation
r ¼ 0.6 (n ¼ 28) and standard error of 1.2 g/dL for
Hgb, resulting from the variability in melanin concentration affecting tissue diffusion, which is not factored in the regression analysis.
Zhang et. al. reported on the utility of a similar
visible/NIR diffuse reflectance spectroscopy technique using the probe pictured in Figure 7. This group
focused more on the ability to monitor inpatient Hgb
Figure 7 Reflectance probe to detect HCT (a) and a close
up of the irradiation, collection fibers in the probe (b). Reproduced with permission from [39].
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
J. McMurdy et al.: In vivo hemoglobin determination
fluctuation during operative procedures, in this case
cardiopulmonary bypass. Collecting both diffuse reflectance spectra and determining Hct through in vitro fingerstick method, a correlation is made to concentration fluctuations during the procedure. Using a
partial least squares regression, this group reports an
intra-patient correlation r ¼ 0.844 (n ¼ 10 patients)
between the fingerstick and diffuse reflectance method, but only an interpatient correlation r ¼ 0.509
when comparing cross patient data using a single PLS
model. This group has reported variations of this
technology with similar results in additional human
patient studies [40] as well as animal models [41].
In both of these cases, correlating diffuse reflectance signals to Hgb or Hct showns a low degree of
correlation as a result of the high variability in optical parameters from one patient to the next. This
technique is likely more suited as it is applied in the
latter study, to track hemoglobin fluctuations dynamically, as the tissue properties other than Hgb are
more consistent in sequential measurement from an
individual patient
One method of minimizing the effects of varying
tissue properties is to collect a diffuse reflectance
spectra from a mucosal surface, as explored in the
imaging systems above. This can have the advantage
of simplifying the device to a single point spectrometer, but making the calibration and regression analysis less susceptible to variation in tissue optical properties. Our group has disclosed the use of visible
reflectance spectroscopy analyzing the palpebral conjunctiva (inner lining of the eyelid) as a method of
noninvasively monitoring Hgb [42, 43]. Diffuse reflectance spectra were collected from the palpebral conjunctiva and analyzed using a partial least squares regression calibration and leave one out cross validation
with the prediction set also used as the calibration set.
A correlation r ¼ 0.92 ðn ¼ 30Þ and standard error of
0.67 g/dL were obtained for Hgb in the initial trial
using a opthalmoscope-type configuration (pictured
in Figure 8(a)), while a second study showed correlation r ¼ 0.82 ðn ¼ 102Þ and standard error of 1.05 g/
dL using a fiber optic reflectance probe (pictured in
Figure 8(b)). Both these data sets were across an eth-
Figure 8 Experimental configurations to collect diffuse reflectance spectra from the palpebral conjunctiva using a
slit-lamp type configuration (a) and fiber optic array reflectance probe (b).
www.biophotonics-journal.org
REVIEW
ARTICLE
J. Biophoton. 2, No. 5 (2009)
nically diverse population to demonstrate the potential advantages of the palpebral conjunctiva over
other tissue segments, reported by physicians as the
optimal physiological location to quantitatively look
for pallor to identify severe anemia [44].
The ideal configuration for a reflection device is
most likely a hybrid of these two classes of device, a
high resolution spectral imaging device implemented
to blood vessels of a mucosal surface. This can minimize background effects while maintaining reliability
of Hgb determination across a diverse group of patients and conditions.
6. Conclusions and future directions
While research has been ongoing in this field for
twenty five years, and intensely over the past ten
years with more cost effective optical component solutions, the actual clinical adoption of one of these
technologies will be a result of two factors, the ability to integrate the device with current standards of
care and the accuracy and reliability of results across
a random patient demographic.
In the current standard of care, devices that can be
implemented as modified pulse oximeters are likely
the strongest candidate for rapid adoption. This may
or may not be a function of the long term performance
of the device, but more in the infrastructure in place
and familiarity of staff and patients with this device.
Particularly if Hgb can be inferred from physiological
variation and not an extrinsically induced event, the
shift from the current pulse oximeter to a higher functioning device will be more straightforward.
Regarding accuracy, blood vessels in mucosal surfaces hold the highest potential as a gateway access
point to monitoring Hgb and other blood parameters. Regardless of how much engineering is integrated into the transmissive type devices, the fundamental limits of interpatient variability in many
optical parameters limit the theoretical performance.
In the case of vessels under a thin mucous layer,
these effects are significantly minimized yielding the
potential for higher performing devices using less advanced instrumentation and signal processing. However, it remains to be seen if quantitatively probing
a new physiological location (retina, sublingual mucosa, palpebral conjunctiva) can overcome the adoption barrier to change the standard of care rather
than embrace current practices.
It is clear that many of these technologies are
ready or close to ready for application and the coming
years will show which is found to be the best solution.
Acknowledgements The authors wish to acknowledge financial support for their own research from the NASA
GSRP Program (NNG05GL57H) and RI Science and
www.biophotonics-journal.org
285
Technology Advisory Committee (STAC) Grant and
thank members of the Display and Photonics Laboratory
for feedback on manuscript preparation.
Conflict of Interest Authors are working with and have a
financial interest in Corum Medical, a company pursing
one technology for measuring anemia noninvasively.
John McMurdy recently
completed his Ph.D. in biomedical
engineering
at
Brown University, USA, under the support of a NASA
Graduate Student Researcher Program Fellowship. The
scope of his research has
been in the clinical testing
of spectroscopic techniques
and the design of compact
and inexpensive optical sensors to non-invasively measure blood components. Dr. McMurdy’s expertise is in
spectroscopic techniques and spectral processing, liquid
crystal optics, and biomedical optical diagnoses methods. He has authored nine peer-reviewed publications
as well as three book chapter contributions. In addition
to a three year NASA GSRP award, he has also been
awarded three consecutive SPIE educational scholarships and an Advanced E-Team Grant from the
National Collegiate Inventors and Innovators Alliance
to foster commercialization of a non-invasive anemia
sensor in third world and impoverished countries.
Gregory Jay is an Associate
Professor in the Department
of Emergency Medicine and
Division of Engineering at
Brown University, USA. He
also serves as an Adjunct
Professor in the Department
of Biomedical Sciences at
the University of Rhode Island. In addition to his academic activities, he brings
over 15 years of experience
as an emergency medicine
physician, and he is currently employed in that capacity
by University Emergency Medicine Foundation, practicing at Rhode Island Hospital and Miriam Hospital in
Providence, Rhode Island. Dr. Jay, who holds an M.D.,
Ph.D. (in Biophysics) and double bachelor degrees (in
engineering and biochemistry) from State University of
New York at Stony Brook, is a member of numerous
professional societies and holds over a dozen issued patents and pending patent applications. He has published
50+ articles in medical and scientific journals and is a
recipient of support form NIH.
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Journal of
BIOPHOTONICS
286
Selim Suner graduated from
Brown Medical School after
completing a Masters degree in Biomedical Engineering, also at Brown University, USA. Dr. Suner,
who is on the staff at Rhode
Island Hospital and the director of Disaster Medicine
and Emergency Preparedness in the Department of
Emergency Medicine, completed the Brown University/Rhode Island Hospital Residency Program in
Emergency Medicine. He is currently an associate professor of Emergency Medicine, Surgery and Engineering at Brown University. Dr. Suner is the leader of the
Rhode Island Disaster Medical Assistance Team
(DMAT), and he has served during multiple disaster
deployments including the responses to Hurricane Katrina, the World Trade Center attacks and the Egypt
Air crash off the shores of Massachusetts. Dr. Suner
also serves as Chairman of Rhode Island Hospital
Emergency Preparedness Committee. He is an international expert in emergency preparedness and disaster
medicine and has given over a 100 lectures related to
disaster management, world-wide.
Gregory Crawford is currently the William K. Warren Foundation Dean of the
College of Science and Professor Physics at Notre
Dame University, USA.
Previously, he was the Dean
of Engineering at Brown
University and formerly the
Richard and Edna Salomon
Assistant Professor and Assistant Professor of Engineering at Brown University. His basic research interests include liquid crystals,
polymers, and their application in display technology,
telecommunications, and medical device technology.
Professor Crawford did his postdoctoral work at the
Naval Research Laboratory in Washington D.C. focusing on electroclinic and ferroelectric materials for fast
switching spatial light modulator applications. He received his B.S. in both Physics and Mathematics and a
Ph.D. from Kent State University where he performed
his doctoral research at the Liquid Crystal Institute and
NSF ALCOM Center. He has over 200 research publications, review articles and book chapters, and holds 14
US patents.
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
J. McMurdy et al.: In vivo hemoglobin determination
References
[1] K. I. Yamakoshi, H. Shimazu, T. Togawa, M. Fukuoka, and H. Ito, Non-invasive measurement of hematocrit by electrical admittance plethysmography technique. Ieee Transactions on Biomedical Engineering
27(3), 156–161 (1980).
[2] G. Zonios, J. Bykowski, and N. Kollias, Skin melanin,
hemoglobin, and light scattering properties can be
quantitatively assessed in vivo using diffuse reflectance spectroscopy. Journal Of Investigative Dermatology 117(6), 1452–1457 (2001).
[3] H. Yang, S. Xie, H. Li, and Z. Lu, Determination of
human skin optical properties in vivo from reflectance
spectroscopic measurements. Chinese Optics Letters
5(3), 181–183 (2007).
[4] T. Shi and C. A. DiMarzio, Multispectral method for
skin imaging: development and validation. Applied
Optics 46(36), 8619–8626 (2007).
[5] G. Zonios and A. Dimou, Modeling diffuse reflectance from semi-infinite turbid media: application to
the study of skin optical properties. Optics Express
14(19), 8661–8674 (2006).
[6] J. M. Steinke and A. P. Shepard, Reflectance measurements of hematocrit and oxyhemoglobin saturation. American Journal of Physiology 253(1), 147–153
(1987).
[7] R. G. Nadeau and W. Groner, The role of a new noninvasive imaging technology in the diagnosis of anemia.
Journal Of Nutrition 131(5), 1610S–1614S (2001).
[8] M. B. Pugh, editor. Stedman’s Medical Dictionary.
Lippincott, Williams and Wilkins, Baltimore, MD,
27th edition, 2000.
[9] R. J. Stoltzfus, Defining iron-deficiency anemia in
public health terms: A time for reflection. Journal Of
Nutrition 131(2), 565S–567S (2001).
[10] C. W. Burt. National Health Care Statistics, 7, 2008.
[11] J. G. Webster, Design of Pulse Oximieters. Institute of
Physics Publishing, Bristol, UK, 1997.
[12] S. J. Barker, J. Curry, S. Morgan, and W. Bauder, New
pulse oximeter measures methemoglobin levels in human volunteers. Anesthesia & Analgesia 102, S5, 2006.
[13] S. Suner, R. A. Partridge, A. Sucov, J. Valente,
K. Chee, and G. D. Jay, Non-invasive pulse co-oximetry for carboxyhemoglobin screening in the emergency department: Identifying occult carbon monoxide toxicity. Journal of Emergency Medicine 34(4),
441–450 (2008).
[14] Mark R. Macknet, Suzanne Norton, Penny L. KimballJones, Richard L. Applegate, Robert D. Martin, and
Martin W. Allard, Noninvasive measurement of continuous hemoglobin concentration via pulse CO-Oximetry. CHEST 132(4, Suppl. S), 493S–494S, OCT (2007).
[15] Alain Berrebi, Orna Amir, Aharon Weinstein, Emmanuel Monashkin, and Paul Holland, A non-invasive evaluation of hemoglobin with a new optical sensor. BLOOD 108 (11, Part 1): 381A, NOV 16 (2006).
[16] J. S. Powell, C. J. Gresens, D. Miller, W. Dummer, and
P. Holland, Non-invasive measurement of hematocrit
to screen potential blood donors. Transfusion, 45(3),
157A–157A (2005). Suppl. S.
www.biophotonics-journal.org
REVIEW
ARTICLE
J. Biophoton. 2, No. 5 (2009)
[17] H. Kanashima, T. Yamane, T. Takubo, T. Kamitani,
and M. Hino, Evaluation of noninvasive hemoglobin
monitoring for hematological disorders. Journal Of
Clinical Laboratory Analysis 19(1), 1–5 (2005).
[18] I. Fine, D. Geva, O. Amir, and E. Monashkin, Clinical
evaluation of non-invasive hemoglobin (hb) testing
device. American Society of Anesthesiologists Annual
Meeting 1, A597 (2004).
[19] M. Rendell, E. Anderson, W. Schlueter, J. Mailliard,
D. Honigs, and R. Rosenthal, Determination of hemoglobin levels in the finger using near infrared spectroscopy. Clinical And Laboratory Haematology 25(2),
93–97 (2003).
[20] K. J. Jeon, S. J. Kim, K. K. Park, J. W. Kim, and
G. Yoon, Noninvasive total hemoglobin measurement.
Journal Of Biomedical Optics 7(1), 45–50 (2002).
[21] T. K. Aldrich, M. Moosikasuwan, S. D. Shah, and
K. S. Deshpande, Length-normalized pulse photoplethysmography: A noninvasive method to measure
blood hemoglobin. Annals Of Biomedical Engineering 30(10), 1291–1298 (2002).
[22] Y. Kinoshita, T. Yamane, T. Takubo, H. Kanashima,
T. Kamitani, N. Tatsumi, and M. Hino, Measurement
of hemoglobin concentrations using the astrim (tm)
noninvasive blood vessel monitoring apparatus. Acta
Haematologica 108(2), 109–110 (2002).
[23] D. Gravenstein, S. Lampotang, N. Gravenstein, and
M. Brooks, Noninvasive hemoglobinometry. Anesthesiology 81(3A), A576–A576 (1994). Suppl. S.
[24] G. Yoon, S. J. Kim, and K. J. Jeon, Robust design of
finger probe in non-invasive total haemoglobin monitor. MEDICAL & BIOLOGICAL ENGINEERING
& COMPUTING 43(1), 121–125 JAN (2005).
[25] R. R. Steuer, D. H. Harris, R. L. Weiss, M. C. Biddulph, and J. M. Conis, Evaluation of a noninvasive
hematocrit monitor: A new technology. American
Clinical Laboratory 10(6), 20–22 (1991).
[26] A. Berrebi and I. Fine, Non-invasive measurement of
hemoglobin/hematocrit over a wide clinical range
using a new optical signal processing method. Blood
94(10), 10B–10B (1999). Part 2 Suppl. 1.
[27] D. Geva, B. Z. Sklar, E. Menashkin, and I. Fine, New
approach in non-invasive measurement of hb/hct/
sp0(2). Anesthesiology 93(3A), U154–U154 (2000).
[28] L. D. Shvartsman and I. Fine, Optical transmission of
blood: Effect of erythrocyte aggregation. Ieee Transactions On Biomedical Engineering 50(8), 1026–1033
(2003).
[29] K. Saigo, S. Imoto, M. Hashimoto, H. Mito, J. Moriya,
T. Chinzei, Y. Kubota, S. Numada, T. Ozawa, and
S. Kumagai, Noninvasive monitoring of hemoglobin –
the effects of wbc counts on measurement. American
Journal of Clinical Pathology 121(1), 51–55 (2004).
[30] J. W. Winkelman, Noninvasive blood bell measurements by imaging of the microcirculation. American
Journal of Clinical Pathology 113(4), 479–483 (2000).
[31] O. Genzel-Boroviczeny, A. G. Harris, G. D. Fletcher,
and F. Christ, Non invasive determination of hemoglobin in neonates with orthogonal polarization spectral (ops) imaging. Pediatric Research 47(4), 332A–
332A (2000). Part 2 Suppl. S.
www.biophotonics-journal.org
287
[32] W. Groner, J. W. Winkelman, A. G. Harris, C. Ince,
G. J. Bouma, K. Messmer, and R. G. Nadeau, Orthogonal polarization spectral imaging: A new method
for study of the microcirculation. Nature Medicine
5(10), 1209–1213 (1999).
[33] M. J. Rice, R. H. Sweat, Jr., J. M. Rioux, W. T. Williams, and W. Routt, Non-invasive measurement of
blood components using retinal imaging, 2002.
[34] J. W. Winkelman. Apparatus and method for in vivo
analysis of red and white blood cells, 1991.
[35] R. Leitgeb, C. K. Hitzenberger, and A. F. Fercher,
Performance of fourier domain vs. time domain optical coherence tomography. Optics Express 11(8),
889–894 (2003).
[36] N. V. Iftimia, D. X. Hammer, C. E. Bigelow, D. I. Rosen, T. Ustun, A. A. Ferrante, D. Vu, and R. D. Ferguson, Toward noninvasive measurement of blood hematocrit using spectral domain low coherence
interferometry and retinal tracking. Optics Express
14(8), 3377–3388 (2006).
[37] X. M. Wu, S. Yeh, T. W. Jeng, and O. S. Khalil, Noninvasive determination of hemoglobin and hematocrit
using a temperature-controlled localized reflectance
tissue photometer. Analytical Biochemistry 287(2),
284–293 (2000).
[38] O. S. Khalil, S. Yeh, M. G. Lowery, X. Wu, C. F. Hanna, S. Kantor, T. W. Jeng, J. S. Kanger, R. A. Bolt, and
F. F. de Mul, Temperature modulation of the visible
and near infrared absorption and scattering coefficients of human skin. Journal of Biomedical Optics
8(2), 191–205 (2003).
[39] B. R. Soller, M. Cabrera, S. M. Smith, and J. P. Sutton,
Smart medical systems with application to nutrition
and fitness in space. Nutrition 18(10), 930–936
(2002).
[40] B. R. Soller, S. Zhang, S. Kaur, S. O. Heard, C. Gaca,
M. J. Rohrer, B. S. Cutler, and T. J. V. Salm, Application
of near infrared spectroscopy for the noninvasive measurement of hematocrit on surgical patients. Critical
Care Medicine 27(12), A144–A144 (1999). Suppl. S.
[41] B. R. Soller, N. Cingo, J. C. Puyana, T. Khan, C. Hsi,
H. Kim, J. Favreau, and S. O. Heard, Simultaneous
measurement of hepatic tissue ph, venous oxygen saturation and hemoglobin by near infrared spectroscopy. Shock 15(2), 106–111 (2001).
[42] J. W. McMurdy, G. D. Jay, S. Suner, F. M. Trespalacios,
and G. P. Crawford, Diffuse reflectance spectra of the
palpebral conjunctiva and its utility as a noninvasive
indicator of total hemoglobin. Journal of Biomedical
Optics 11(1), 014–019 (2006).
[43] G. D. Jay, J. Racht, J. W. McMurdy, Z. Mathews,
A. Hughes, S. Suner, and G. P . Crawford, Point-ofcare noninvasive hemoglobin determination using fiber optic reflectance spectroscopy. Proceedings of
IEEE: Engineering in Medicine and Biology Society
1, 2932–5, 2007.
[44] O. L. Hung, N. S. Kwon, A. E. Cole, G. R. Dacpano,
T. Wu, W. K. Chiang, and L. R. Goldfrank, Evaluation of the physician’s ability to recognize the presence or absence of anemia, fever, and jaundice. Academic Emergency Medicine 7(2), 146–156 (2000).
# 2009 by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim