Research Article Vol. 55, No.

31 / November 1 2016 / Applied Optics 8951

Determining the refractive index of human

hemoglobin solutions by Kramers–Kronig
relations with an improved absorption model
JONAS GIENGER,* HERMANN GROß, JÖRG NEUKAMMER, AND MARKUS BÄR
Physikalisch-Technische Bundesanstalt (PTB), Abbestraße 2–12, 10587 Berlin, Germany
*Corresponding author: [email protected]

Received 22 July 2016; revised 21 September 2016; accepted 27 September 2016; posted 7 October 2016 (Doc. ID 272064);
published 31 October 2016

The real part of the refractive index of aqueous solutions of human hemoglobin is computed from their absorp-
tion spectra in the wavelength range 250–1100 nm using the Kramers–Kronig (KK) relations, and the corre-
sponding uncertainty analysis is provided. The strong ultraviolet (UV) and infrared absorbance of the water
outside this spectral range were taken into account in a previous study employing KK relations. We improve
these results by including the concentration dependence of the water absorbance as well as by modeling the deep
UV absorbance of hemoglobin’s peptide backbone. The two free parameters of the model for the deep UV
absorbance are fixed by a global fit. © 2016 Optical Society of America
OCIS codes: (000.1430) Biology and medicine; (000.3860) Mathematical methods in physics; (120.4530) Optical constants;
(260.2030) Dispersion; (300.1030) Absorption.

http://dx.doi.org/10.1364/AO.55.008951

1. INTRODUCTION knowledge of the Hb RI is of crucial importance to reliably

The optical properties of biological cells and tissues have been calculate RBC volume and Hb content.
subject to research for many decades. Refractometry in cells is Within the blood of a single person, the intracellular Hb
used, for example, for protein detection [1]. Recently, the re- concentration fluctuates among the individual RBCs, with a
fractive index (RI) or RI distribution of cells has been measured coefficient of variation around 6%–8% [5,6]. Because of the
by phase microscopy [2], holographic techniques [3], absorp- significant influence of this variation, it does not suffice to
tion cytometry [4], or optical tomography [5]. From such know the RI of Hb solutions at the mean concentration, but the
measurements, the dry mass or concentrations of proteins in dependence of the RI on the concentration has to be specified.
The absorption spectra of Hb solutions were measured with
the cell can be derived, provided the relation to the optical
properties is known. high accuracy over a wide range of wavelengths and have been
Analysis of blood samples includes the determination of the known for several decades. In contrast, measurements of their
quantities of the so-called complete blood count (CBC), one RI, especially at physiologically relevant high concentrations,
of the most frequently performed measurements in laboratory are challenging and have only been presented as late as 2005
medicine. Beside the concentrations of red blood cells (RBCs, for a wide spectral range [8]. However, these data have much
erythrocytes), white blood cells, and platelets, an important in- larger measurement uncertainties than the corresponding ab-
dicator for diseases (e.g., anemia) is the mean cellular volume sorption spectra, such that the real part of the complex RI is
(MCV) of RBCs. Beside the calculation of the MCV as the less accurately known than its imaginary part, the absorbance.
ratio of the hematocrit value and the RBC concentration, As a remedy, several authors [9,10] suggested employing
flow-cytometric detection of light scattering by RBCs has been Kramers–Kronig (KK) relations in order to obtain the real part
used for more than three decades to determine the volume and of the RI directly from accurate measurements of absorption
hemoglobin content of individual cells [6,7] at a throughput in spectra (i.e., the imaginary part of the RI as a function of wave-
the range of 1000 events per second. Of course, a detailed length). However, this is a nontrivial task, due to the finite
knowledge of the cells’ RI is required for this. Since RBCs wavelength range of the measured spectra and the global,
are mainly composed of the protein complex hemoglobin long-range character of the KK relations that yield the real part
(Hb), which is dissolved in water at an average intracellular of the RI as an integral transformation of the imaginary part. In
concentration of typically 340 gL−1 for healthy persons, the this paper, we supplement and extend the literature spectra [8]

1559-128X/16/318951-11 Journal © 2016 Optical Society of America

 8952 Vol. 55, No. 31 / November 1 2016 / Applied Optics Research Article

with an absorption model for proteins toward the UV. As a ϵM
λ, where λ is the vacuum wavelength of the light. The
result, we obtain RIs that are in better agreement with results latter implies that the Lambert–Beer law holds, which is the
obtained by reflectance measurements for oxyhemoglobin case for Hb at physiological (c ≈ 300 g L−1 ) or lower concen-
solutions [11]. Furthermore, we obtain results for the RI of trations [8]. Instead of μa
λ or ϵM
λ, we express the absorp-
deoxyhemoglobin and give the refractive increment (i.e., the tion spectra in terms of the imaginary part of the complex RI
slope of the RI with respect to concentration) for the first time. n
λ  n
λ  iκ
λ (1)
We also extend the previous treatment by considering the mea-
surement uncertainties of the obtained RIs. These uncertainties of the solution. Here n, κ are positive real functions. The
result from the propagation of the uncertainties of the mea- conversion rule is
sured absorption spectra [8] and the RI data obtained by reflec- μa
λλ ln 10ϵM
λcλ
tance measurements [11], to which we fit the two free κ
λ   : (2)
4π 4πM
parameters of the deep UV absorption model. The mathemati-
The data on the molar extinction coefficient ϵM
λ from
cal model for the absorption spectra and related method are
Ref. [14] are converted to κ
λ, using the molar mass of the
presented in Sections 2 and 3. Our approach can be applied
Hb tetramer M  64458 g mol−1 [17].
to any variant of Hb. Section 2A describes the imaginary part
of the RI of a Hb solution in dependence on concentration. KK 2. Total Absorbance of Hb Solutions and Erythrocytes
relations are used in Section 2B to obtain an expression for the In the spectral range of λ ∈ 250; 1100 nm, the strongest
real part of the solution’s RI. A model for the deep UV absorb- absorption of light by RBCs and hence, of blood, is caused
ance of Hb is proposed in Section 2B as well. Section 2C de- by Hb. Water has a fairly low absorption coefficient in this re-
scribes the fitting of the two free-model parameters to the gion (cf. Fig. 1) and other RBC components contained in the
literature data for the real part of the RI. In Section 3, we apply cytosol (e.g., other proteins, sugars, ions) exhibit rather low con-
the analysis to experimental data for oxygenated and deoxygen- centrations. We thus consider a two-component system, i.e.,
ated hemoglobin, respectively. The significance of the results is
discussed, and a comparison to previous KK-analyses is made. 1. The solvent, or simply “Water” (H2 O), occupying a
In Section 4, we summarize our findings. volume V H2 O and
2. The absorbing solute, or simply “Hemoglobin” (Hb),
A. Terminology denoting everything else contained in the erythrocyte cytosol
Oxygenated hemoglobin is often abbreviated as “HbO2 ” while and occupying a volume V Hb . The solute is treated as pure
deoxygenated hemoglobin is usually abbreviated as “Hb.” In Hb for its other physical properties, such as molar mass and
order to simplify the notation, we will use the term “Hb” density.
for hemoglobin in general, specifying the hemoglobin variant Let the volume fraction occupied by Hb molecules be ϕ 
only if relevant. V Hb ∕
V H2 O  V Hb  and the volume fraction occupied by
The RI of Hb solutions is complex-valued. For the sake of water molecules 1 − ϕ  V H2 O ∕
V H2 O  V Hb . Since the
brevity, we refer to the real part of the RI by “real RI” and to the
imaginary part of the RI by “imaginary RI.” This shall not
imply that the RI is purely real or imaginary.

2. MATERIALS AND METHODS

A. Absorption Spectra: Imaginary Part of RI
The Hb absorption spectra in the wavelength range [250,1100]
nm reported in Ref. [8] were used in our calculations. These
data were measured for Hb solutions produced from human
erythrocytes by repeated freezing and thawing followed by cen-
trifugation of the membrane components. These homogeneous
solutions did thus contain all the constituents of human eryth-
rocytes, except for the membranes (thickness less than a few
10 nm, volume fraction less than a few percent [6,12,13]).
We further use experimental data for pure Hb solutions to
supplement the absorbance data in the wavelength range
228–250 nm [14]. Fig. 1. Imaginary refractive increment α
λ of Hb in aqueous solu-
Other researchers [9,10] have previously relied on the data tions (left axis, solid and dashed–dotted lines). Experimental data are
compiled from various sources by Prahl [15]. However, since taken from Ref. [8] (corrected for H2 O absorbance) and Ref. [14].
not all of the compiled sources used human Hb, these data are A Lorentzian peak models the deep UV absorbance of the peptide back-
not used here. bone (left axis, solid black line). Imaginary RI κH2 O
λ of water (right
axis, dashed line) [16]. Note that the quantities α; κH2 O have different
1. Literature Data units and scaling of y-axes. To compare the numerical values of α and
The literature absorption spectra we use in our calculations are κH2 O , one needs to multiply α by the respective concentration of the
expressed in terms of the inverse absorption length μa
λ at a RBC (e.g., c Hb  340 gL−1 ) [cf. Eq. (5)]. Spline interpolation was
given mass concentration c, or the molar extinction coefficient applied to obtain a step width of 1 nm.
 Research Article Vol. 55, No. 31 / November 1 2016 / Applied Optics 8953

Lambert–Beer law holds, we make the following ansatz for the B. Real Part of RI by KK Relations
total imaginary RI (absorbance) of the solution: 1. Literature Data
The most complete experimental measurement of the wave-
κ
λ  ϕκ Hb
λ 
1 − ϕκ H2 O
λ; (3) length-dependent real RI of Hb solutions to date was presented
in Refs. [8,11]. Here, n
λ was determined via measurements of
where κH2 O
λ is the absorbance of pure water and κ Hb
λ is
the spectral reflectance R
λ at an interface between air and Hb
the absorbance of “pure Hb in aqueous solution.” It should be
solution at normal incidence. The reflectance at normal inci-
noted that κ Hb is not equal to the imaginary RI of a crystal of
dence is connected to the complex RI by the Fresnel equation
pure Hb, as is revealed by comparison with data for the com-  
 n
λ − 1 2
 
n
λ − 1  κ
λ ;
2 2
plex dielectric function of thin Hb films [18]. Equation (3) is R
λ    (8)
similar to the ansatz in Ref. [10], where, however, the authors n
λ  1
n
λ  12  κ
λ2
did not include the prefactor 1 − ϕ for the water absorbance, which is easily solved for n
λ when κ
λ is known or when it
which results in a relevant difference, as shown later. This ex- can be neglected because κ
λ ≪ n
λ − 1.
cluded volume effect (i.e., the fact that there is less water in a In Ref. [11], measurements at different concentrations were
solution of higher Hb concentration) is essential and must be analyzed. A linear-affine dependence of the real RI on the con-
taken into account, since the volume fractions of Hb can be as centration was found and the result was expressed as
high as 26% or more.
The Hb volume fraction is related to the concentration by n
λ  nH2 O
λ1  c Hb β
λ: (9)
c Hb Here β
λ is the concentration-specific increment of the real RI
ϕ ; (4) relative to the water RI.
ρHb
2. KK Relations
where ρHb is the mass density of a hypothetical solution con-
The complex RI is a linear, causal response-function to an in-
taining 100% Hb and 0% H2 O. Both κ Hb
λ and ρHb are co-
cident wave. Hence, its real and imaginary part are related by
efficients in a linear interpolation of spectra and density that
KK relations. Expressed in terms of wavelengths, the KK rela-
holds at least up to physiologically high concentrations. A value
tions for the complex RI read
of ρHb  1330 gL−1 for proteins is given in Ref. [1], indepen- Z
dent of the type of protein. A normal physiological intraery- 2 ∞λ λ
n
λ − 1  Kκ
λ≔ − κ
ΛdΛ; (10)
throcyte Hb concentration of 340 gL−1 thus corresponds to a π 0 Λ Λ − λ2
2

volume fraction of ϕ ≈ 0.26 or 26%.

Then the term for the absorbance contribution of Hb, Z
2 ∞ λ
ϕκ Hb
λ, becomes −1
κ
λ  K n
λ   n
ΛdΛ; (11)
π 0 Λ2 − λ 2
κHb
λ def :α
λ where Eq. (10) also defines the integral transform K in
ϕκ Hb
λ  c Hb  c Hb α
λ; (5) R general.
ρHb The symbol ≔ denotes a definition and the symbol denotes
the Cauchy principal value integral.
where α
λ is Hb’s concentration-specific increment of the Applying Eq. (10) formally to the ansatz for the absorption
imaginary RI, or the imaginary refractive increment. Since of the Hb solution to Eq. (6), we obtain
the Hb concentration c Hb is measured in g L−1 , the unit for

α is L g−1 . Equation (3) becomes c
n
λ − 1  c Hb G
λ  1 − Hb
nH2 O
λ − 1; (12)

 ρHb
c
κ
λ  c Hb α
λ  1 − Hb κ H2 O
λ: (6) where G
λ≔Kα
λ is the transformed spectrum [cf.
ρHb Eq. (10)]. This formal transformation of the absorption results
Equation (6) together with Eqs. (2) and (3) allows one to in an equation for the real RI of the Hb solution
 
compute the imaginary refractive increment α
λ from a mea- nH2 O
λ − 1
n
λ  nH2 O
λ  c Hb G
λ −
surement of the inverse absorption length μa
λ of a solution at ρHb
known concentration c Hb as def :B
λ
 nH2 O
λ  c Hb B
λ: (13)
1  
α
λ  κ
λ 
ϕ − 1κ H2 O
λ The linear-affine dependence of n
λ on c Hb in Eq. (13) is in
c Hb

   agreement with experimental findings [11] [cf. Eq. (9)]. In
1 λ 
c Hb Eq. (6), we have formally split off the water absorption, such
  μ
λ  − 1 μa;H2 O
λ ; (7)
c Hb 4π a ρHb that nH2 O
λ contributes to the background in the dispersion
relations of the Hb solutions [cf. Eq. (13)]. Since the real RI of
where the asterisk  denotes experimental data. The inverse water, unlike the real RI of Hb, is known to high accuracy, this
absorption length of water μa;H2 O
λ is known to high accuracy provides valuable additional information compared to the ap-
over a large spectral range λ ∈ 10 nm; 10 m [16]. Hence, plication of the KK-transform only to the measured absorption
this formula allows one to correct for the water absorption, spectrum of a Hb solution in the visible and near UV/IR range
which is important in the infrared (IR), where Hb absorbs that was presented in Ref. [9]. This idea was already presented
only weakly. in Ref. [10]. However, due to our different ansatz for the
 8954 Vol. 55, No. 31 / November 1 2016 / Applied Optics Research Article

absorption, where we take into account the excluded water vol- As mentioned before, the deep UV spectrum, and hence, the
ume, we obtain a different result for B
λ with an additional term exact shape of the peptide absorption line, is not available in

nH2 O
λ − 1∕ρHb for the concentration dependence. We dis- the literature. For proteins, Woods and O’Bar report that “the
cuss the differences between the results obtained in Refs. [9,10] increase in absorbance at 187 nm is threefold over that at
and the result with our improved model in Section 3B. 205 nm and fourfold over that at 210 nm” [21]. This description
For numerical values of nH2 O
λ, we use a four-term Sellmeier fits well to the half-width of the curve of Γ  11.6 nm. Going
formula that is accurate to at least five decimal places [19]. to even lower wavelengths λ ≪ 187 nm, there will be more ab-
sorption features, since the inner electron shells of the atoms will
3. Additional Spectral Information outside the Measured be excited. We thus expect a variety of overlapping absorption
Range lines at these short wavelengths. On the other hand, this spectral
The KK relations provide a formal tool to derive results like region is fairly far away from the region of interest and the KK
Eq. (13). However, their application has a well-known problem relations contain a damping factor of 1∕
λ2 − Λ2 . Hence, the
for the numerical evaluation: They are global integral transforms exact line-shapes are practically irrelevant. For our model, it suf-
that require the knowledge of real or imaginary RI at all wave- fices to add to the spectrum a delta-peak of unknown amplitude
lengths λ ∈ 0; ∞, which is practically impossible. Although the located at zero wavelength, which accounts for the influence of
integral kernel in Eq. (10) is decaying with increasing distance extreme UV absorption by a constant offset in the real
from the pole at Λ  λ, it is long-ranged. Hence, one cannot RI: αδ
λ  limλδ →0 π2 aδ λδ δ
λ − λδ .
simply cut off the integration domain (i.e., use a finite data set). In this respect, our approach is not different from
Water is transparent to visible light, and its main regions of Refs. [9,10]: We cannot predict the absolute value of the RI
absorption lie in the UV and the IR (Fig. 1). Due to these ab- of Hb, but we need to determine a constant by comparing
sorption bands, water has a RI significantly different from 1, to experimental data. However, we apply nonlocal fitting to
even in the transparent regions, and exhibits normal dispersion optimize this constant, instead of using just one single data
[i.e., n
λ decreases with λ]. This implies that neglecting such point. Thus, our method is more robust to uncertainties in
contributions in the KK relations may lead to inaccurate results. both the absorption spectra and the measured real RIs.
We will now describe the absorption features of Hb below
250 nm by a mathematical model and include them in our C. Fitting to Literature Values
expression for the real RI. With this model for the UV absorption, we have an absorption
In addition to the known absorption spectrum (Fig. 1) with spectrum
strong absorption in the vicinity of the Soret band at 420 nm,
Hb has an even stronger absorption peak in the deep UV. This α
λ  αlit
λ  αL
λ  αδ
λ; (15)
feature stems from the peptide bonds forming the backbone of where αlit
λ represents the experimental literature data. For
any polypeptide or protein, including the protein complex Hb the integral transform K [cf. Eq. (10)], the contribution from
and is characteristic for polypeptides and proteins. The corre- the δ-peak αδ
λ is G δ
λ  aδ and hence constant and the
sponding extinction coefficient curves ϵ
λ are similar among a contribution G L
λ  aL G̃ L
λ from the Lorentzian, inte-
variety of proteins and the absorbance maximum is typically grated only over the deep UV part of the spectrum, can be ob-
located at λ  187 nm [20,21]. tained analytically (see Appendix B). Here G̃ L
λ is the
This peptide-peak must be accounted for to perform a proper contribution for a Lorentzian of unit amplitude. Thus, we
KK analysis, but is, unfortunately, not resolved in the existing are left with
experimental Hb spectra. However, data were reported for hu-
man and bovine albumin [21], a protein found in blood serum. G
λ  G lit
λ  aL G̃ L
λ  aδ ; (16)
Albumin is similar to Hb in its mass and optical properties at
where only the first term G lit
λ≔Kαlit 
λ needs to be evalu-
wavelengths away from the characteristic Hb absorption band
ated numerically.
at 420 nm. The absorption maximum for human albumin is
Numerical evaluation is straightforward. We use an integra-
reported as ϵ
187 nm  86.0 L g−1 cm−1 , corresponding to
tion scheme that evaluates the KK relations as a Riemann sum
a value of α
187 nm  2.95 × 10−4 L g−1 , which is more than
with Taylor expansion at the singularities of the integrand as de-
four times as high as the peak around 420 nm (Fig. 1).
scribed (e.g., in Ref. [22] and in Appendix A). The experimental
We model this generic protein absorption using an antisym-
literature data αlit
λ comprises two data sets: (1) the absorbance
metrized Lorentzian curve
data in the range 228–250 nm [14], which we refer to as “ultra-
1 Γ 1 Γ violet” (UV), and (2) the data in the range 250–1100 nm [8],
αL
λ  aL −a ; (14)
π
λ − L2  Γ2 L π
λ  L2  Γ2 referred to here as “visible” (VIS). Although our use of the terms
where Γ  11.6 nm is the half-width at half-maximum of the “visible” and “ultraviolet” deviates from the usual definition, it is
curve, and L  187 nm is the position of its maximum. This convenient to distinguish between the two data sets:
model curve fulfills αL
−λ  −αL
λ. This is important, as the (
αUV
λ  data from14 λ ∈ 228;250 nm;
symmetries κ
−λ  −κ
λ and n
−λ  n
λ are implied αlit
λ  αVIS
λ  data from8 λ ∈ 250;1100 nm; (17)
when using the KK relations in the form Eqs. (10) and 0 else:
(11), or the analogous expressions for the frequency, where they
are written
R as a semi-infinite integral, denoted by the sym- The numerical KK transform is applied to both parts sep-
bol ∞0 . arately, such that G lit
λ  G UV
λ  G VIS
λ.
 Research Article Vol. 55, No. 31 / November 1 2016 / Applied Optics 8955

1. Fitting of Free Parameters fˆ  Hâ  H
HT V−1 H−1 HT V−1 y; (25)

Neither of the two free parameters of the model, aL and aδ , can |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
≕F
be computed from literature data a priori with satisfying accu-
racy. For the peptide absorption aL , the order of magnitude can for parameter and function vector, and B̂  B 0  fˆ for the
be estimated from the semiquantitative data [21], where the refractive increment. The resulting amplitude of the Lorentzian
absorbance maximum is given. It is important to keep in mind is shown in Fig. 1 as generic peptide absorbance.
that the KK transform of the peptide-peak in the deep UV de-
pends much more strongly on the center position and the area 2. Deoxyhemoglobin
under the peak than on its actual maximum. Since the peak Both the absorption spectra of oxygenated and deoxygenated he-
shape is not quantitatively known, the peak height does not moglobin are well known (cf. Fig. 1). However, the method of
contain enough information to determine aL . Ref. [8] to measure the real RI could only be applied to oxy-
The real RI, hemoglobin, whereas measurements with deoxyhemoglobin
were not possible due to precipitation resulting in backscattering
n
λ; aL ; aδ   nH2 O
λ  c Hb B
λ; aL ; aδ ; (18)
of light from highly concentrated solutions of deoxyhemoglobin.
and the real refractive increment, Nevertheless, the KK analysis allows us to derive a result for the
real refractive increment of deoxyhemoglobin. To this end, we
nH O
λ − 1 use the same model for the deep UV absorbance as above,
B
λ; aL ; aδ   G lit
λ− 2  aL G̃ L
λ  aδ ; (19)
ρHb Lorentzian and delta peak, with the coefficients âL , âδ found
|fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl
ffl}
≕G H2 O
λ by the least-squares fit for oxyhemoglobin. Combining this with
the KK transform of the spectrum αdeoxy
λ for deoxyhemoglo-
linearly depend on the parameters aL and aδ . Thus we use a bin, we calculate the real refractive increment according to
linear least-squares approach to optimize the parameter values.

λi   fˆ
λi 
deoxy
The empirical model function in Ref. [11] is formally iden- B deoxy
λi   B 0
tical to ours, but the quantity β
λ  B
λ∕nH2 O
λ was con-

λi   G H2 O
λi   fˆ
λi :
deoxy
sidered instead of B
λ [cf. Eqs. (9) and (13)]. The  G lit (26)
measurements are given at wavelengths λi , i  1; …; N , and The reason to simply use the same deep UV model curve for oxy-
we denote these experimental values by β
λi . We convert hemoglobin as well as for deoxyhemoglobin is that oxygenation
them to B i  β
λi nH2 O
λi . In the following, the B i will affects the heme groups in the Hb complex. The absorption peaks
be referred to simply as “the measurement data.” at about 420 and 560 nm are altered by oxygenation (cf. Fig. 1)
At each wavelength, the increment B i ≔B
λi  consists of a but not the peptide backbone causing the deep UV absorption.
fixed part resulting from numerical KK transformation,
D. Uncertainties
B 0;i  G lit
λi   G H2 O
λi ; (20) Both, the KK transformation and the linear least-squares fit to
and a function yet to be determined, which models the deep the data are linear transformations, which can formally be car-
UV contributions ried out by matrix multiplication. Hence, it is easy to perform
the uncertainty propagation in terms of mean values and covari-
def :h
λ X
f i  aL G̃
λi   aδ r ar hr
λi : (21) ance matrices. A detailed description is given in Appendix C.
rL;δ All values for uncertainties given here correspond to one stan-
dard deviation.
Switching to matrix-vector notation, we write this function vec- An important uncertainty contribution stems from the un-
tor as f  Ha, with H≔fhr
λi gir ∈ RN ×2 , and a 
aL ; aδ T certainty of the Hb concentration c Hb . Because this influences
is the parameter vector. the concentration-specific transmittance and reflectance spectra
Now we want to minimize the deviation between KK results
by a global factor, the uncertainties of the resulting real refrac-
with deep UV absorbance model B and measurement data B  ,
tive increment are strongly correlated among all wavelengths,
B − B  y − f ; (22) i.e., the correlation coefficient
where y  B− B 0 is the data vector. The entries of vector B 0 cov
B i ; B j 
r ij ≔ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; (27)
are given in Eq. (20). The linear least-squares problem is then var
B i var
B j 
χ 2
a → min with
is close to 1 even if λi − λj is large. Here “cov” and “var” de-
χ 2
a≔
y −f T W
y −f  note covariance and variance, respectively. In such a case, the
Xnm
1100 use of the diagonal elements of the covariance matrix only (i.e.,
 wij B 
λi −B
λi ;aB 
λj −B
λj ;a; (23) the variances) may lead to unnecessarily less significant results
λi 250 nm with respect to the uncertainty estimation. When, for instance,
where W  fwij gN is a weight matrix given by the inverse of the difference between two random variables X , Y is consid-
i;j1
the covariance matrix V of the data vector. The conditions for ered, the variance is
minimal χ 2
a are solved by standard linear algebra, which yields var
X − Y   var
X   var
Y  − 2 cov
X ; Y ; (28)
−1 −1 −1
â  arg min χ
a 
H V H H V y;
2 T T
(24) which may be small even if var
X ; var
Y  are large.
 8956 Vol. 55, No. 31 / November 1 2016 / Applied Optics Research Article

3. RESULTS AND DISCUSSION related to its shape, described equivalently by the derivative
A. Real Part of RI for Oxy- and Deoxyhemoglobin B̂ 0
λ, are smaller.
Figure 2 shows the result of the presented KK analysis for the real The result B̂
λ of the KK analysis for oxyhemoglobin and
refractive increment B̂
λ for oxyhemoglobin along with the ex- the experimental data B 
λ as shown in Fig. 2 have a similar
perimental data B 
λ from Ref. [11]. The uncertainties are overall shape, but there are deviations that cannot be accounted
shown as shaded bands, where the half-width of the band cor- for by the measurement uncertainties of B 
λ given in
responds to one standard deviation derived by covariance-matrix Ref. [11] and the uncertainties of the KK result derived, as de-
calculus. The analogue result for deoxygenated hemoglobin scribed in Appendix C. For instance, the difference between the
is shown in Fig. 3. Figure 4 shows the derivative peak at 433 nm and the dip at 399 nm for B  is
B̂ 0
λ  d B̂
λ∕dλ. Here the strong correlation is eliminated, B 
433 nm − B 
399 nm 
7.5
0.6 × 10−5 L g−1 ;
resulting in much smaller uncertainties. It is evident from
while that for our result is
Fig. 4 that the main uncertainty of the result concerns the exact
vertical position of the curve B̂
λ, whereas the uncertainties B̂
433 nm − B̂
399 nm 
5.47
0.06 × 10−5 L g−1 :
The small value for the uncertainty of B̂
433 nm −
B̂
399 nm is due to the fact that the strongly correlated con-
centration contributions to the uncertainty cancel out almost
completely for this difference. This observation also holds for
the derivative B̂ 0
λ (cf. Fig. 4). The difference between the two
values clearly exceeds the uncertainties.
The deviations between B̂
λ and B 
λ cannot be attributed
to unknown spectral absorptions outside the 250–1100 nm
range, since features producing such discrepancies would neces-
sarily be inside this wavelength range. We have modeled impor-
tant spectral absorptions at the UV end of the spectrum.
Concerning possible IR absorptions that are not considered in
the proposed model, we give the following notes: The absorption
spectra of aqueous hemoglobin solutions in the IR between
1.1 μm and 2.6 μm are dominated by water [8], indicating that
the imaginary refractive increment of Hb is very low in this re-
gion. Hypothetical absorption lines due to Hb at even longer
Fig. 2. Real refractive increment B
λ of aqueous solutions of oxy- wavelengths would contribute to the real RI increment B
λ
hemoglobin: Experimental data [11] and results from this work fitted in the form of constants or functions with a gentle negative slope.
to the data. The shaded bands indicate the measurement uncertainties Any possible influence of the long-wavelength end of the spec-
given in Ref. [11] and computed in Appendix C, respectively. trum of Hb would thus change the agreement between our result
B̂
λ and the literature data B 
λ for the worse. Thus we

Fig. 3. Real refractive increment of aqueous hemoglobin solutions: Fig. 4. Derivative of the real refractive increment B 0
λ  ddλ B
λ:
Result for deoxyhemoglobin obtained by KK relations with the same Comparison between results for deoxyhemoglobin and oxyhemoglo-
model for deep UV absorbance as for oxyhemoglobin [cf. Fig. 2]. bin. Estimated standard deviations (shaded bands, barely visible) of
Estimated standard deviations (shaded bands) are larger for deoxyhe- B 0
λ are much smaller than for B
λ (Fig. 3). This is because most
moglobin because B deoxy
λ is composed of more terms with uncorre- of the (strongly correlated) concentration uncertainty cancels out in
lated concentration uncertainties than B̂
λ. the derivative.
 Research Article Vol. 55, No. 31 / November 1 2016 / Applied Optics 8957

conclude that no important contribution to the absorption spec- n
λ  1  c Hb G VIS
λ; (32)
trum was missed at the long-wavelength end. The KK relations which is off the true value by a significant amount. One can
themselves are valid as long as the framework within which the remove this discrepancy by replacing the 1 in the above expres-
data were measured holds (i.e., classical electrodynamics and lin- sion by a free parameter, which can be interpreted as deep UV
ear, causal media). Optical activity of Hb has been examined absorption. Again, this free parameter can be fixed by a single
previously [14,23] and can be ruled out as well, since the ex- measurement at λ0 . The result is then the same as in Eq. (31).
pected effect is too weak. Furthermore, we can exclude biological However, the subtractive KK transform Kα
λ − Kα
λ0  can
variability, since absorption and reflectance/refraction data were be rewritten into a single integral where the kernel decays faster
measured on the same type of sample [8,11]. This implies that than in the standard KK relations, which is numerically favor-
the absorption and refraction data presented in Refs. [8,11] are able and thus given as an argument for the use of subtractive
not self-consistent with the given measurement uncertainties. relations. Thus, the subtractive form of KK relations masks the
Lastly, we address the question of whether these discrepancies lack of knowledge of the absorption spectrum outside the mea-
are of practical importance. At λ  399 nm, where the discrep- sured spectral range: At least one important absorption peak at
ancy is largest, one has B   2.24 × 10−4 L g−1 from Ref. [11] short wavelengths has apparently been omitted. As long as the
and B̂  2.41 × 10−4 L g−1 for our analysis. At a concentration missing peak is far away from the region of interest, the simple
of c Hb  340 g L−1 , the resulting real RIs are n  addition of a constant works fine. However, if the location of
nH2 O  c Hb B   1.420 and n̂  nH2 O  c Hb B̂  1.426, re- the peak becomes important, as is the case for water, this model
spectively, and the residual is jn̂ − n j  0.006  0.4%n . is insufficient.
Since this value is in the subpercent region, it may, at first glance, In Ref. [10], the authors made the ansatz
seem unimportant whether one uses n or n̂. However, when
RBCs are examined with optical methods such as phase-contrast κ 10
λ  κH2 O
λ  c Hb αVIS
λ; (33)
microscopy or light scattering in a flow cytometer, they are usu- for the imaginary RI of the Hb solution, which takes the absorp-
ally immersed in water with an RI of nH2 O
399 nm  1.344, tion due to water into account but neglects the excluded volume
or a saline solution of slightly higher RI. Hence the complex effect for water, in contrast to our approach in Eq. (6). Apart
contrast between cell and surrounding medium is from this difference, the formal application of the KK relations

n 0.061  0.008i 11 in the present work is identical to that in Ref. [10]. The result
m − 1≔ −1 : (29) was
nH2 O 0.057  0.008i our result
n10
λ  nH2 O
λ  c Hb G VIS
λ; (34)
In phase-contrast microscopy, the signal for the optical thickness
of the cell is directly proportional to the contrast m − 1. It is also which provides also a theoretical derivation of the empirical finding
a parameter governing the light scattering by cells. The Rayleigh– n
λ  nH2 O
λ1  c Hb β
λ reported in Ref. [11]. However,
Gans–Debye theory is an approximate description of light scat- the result that β
λ  G VIS
λ∕nH2 O
λ is incomplete, as we
tering for the limiting case of particles with low contrast and have discussed. Subtractive KK relations were used as well to
moderate size. It predicts a scattered electric field proportional match the RI at λ0  800 nm.
to m − 1. The real parts of the above two values for m − 1 differ To compare these two previously presented methods with
by more than 7%, indicating that any quantitative analysis of the ours, we applied them to the spectra of oxyhemoglobin presented
interaction of RBCs with light that requires a priori knowledge of in Ref. [8] and shown in Fig. 1. As an example, a concentration
the contrast cannot be more accurate than this. of c Hb  287 g L−1 was assumed and n
λ0  800 nm 
1.403 was taken from Ref. [8] as well. The comparison of
B. Comparison to Previous KK Analyses
the two methods applied in Refs. [9,10] with our method
We will now briefly review the previous investigations employ- and the experimental dispersion curve given in Ref. [11] is shown
ing KK relations [9,10] and compare the methods. In Ref. [9], in Fig. 5. Neglecting the water influence, as in Ref. [9], yields a
the authors started from Eq. (10) and applied it to the Hb ab- dispersion curve that substantially deviates from the measure-
sorption spectra in a finite spectral range [i.e., instead of ment (dash-dotted black line in Fig. 5) everywhere, except at
Eq. (6)] they assumed and above λ0 , where it was matched. With the water absorption

α
λ λ ∈ a; b and the resulting dispersion taken into account [10], the agree-
κ9
λ  c Hb αVIS
λ  c Hb ; (30)
0 else ment with the experimental result is already much better.
However, the result from Ref. [10] and experimental curve sub-
where a; b is the measured spectral range (i.e., here
stantially differ in the UV below 375 nm. The agreement in the
a; b  250; 1100 nm). The authors then used a subtractive
UV is much better in our approach. This is also evident from
form of the KK relations, where the difference n
λ − n
λ0  is
Fig. 6, where the residuals between KK computations and ex-
considered, which yields
periment are plotted.
n9
λ  n
λ0   c Hb G VIS
λ − G VIS
λ0 : (31)
Here, G VIS
λ≔KαVIS 
λ is the dispersion resulting from the
measured spectrum. The free parameter n
λ0  is fixed by a re- 4. SUMMARY
fractometric measurement at wavelength λ0  800 nm. If the The complex RI of a Hb solution, which forms the cytoplasm
nonsubtractive KK relations had been used instead, the result of erythrocytes and determines their optical properties can be
would have been computed as
 8958 Vol. 55, No. 31 / November 1 2016 / Applied Optics Research Article

for B
λ, Eq. (13). The absorption spectra available in the lit-
erature [8,14] do not resolve the strong UV absorbance of Hb’s
peptide-backbone. Hence, we modeled it by a Lorentzian line
of unknown amplitude, which introduces a free parameter aL
into the expression for B
λ. A second free parameter aδ is in-
troduced as a wavelength-independent term accounting for ex-
treme UV absorbance [cf. Eq. (19)]. These two free parameters
were determined by a linear least-squares fit to the literature
data B 
λ [11]; the result of the fit is denoted by B̂
λ.
We evaluated spectra for oxygenated and deoxygenated he-
moglobin. Data files of the results are provided as supplementary
material for B̂
λ (Data File 1), B deoxy
λ (Data File 2), the cor-
responding covariance matrices (Data File 3 and Data File 4), the
converted literature data for α
λ and αdeoxy
λ [8] and the values
for nH2 O
λ computed according to Ref. [19] (Data File 1 and
Fig. 5. Comparison between the experimental real RI derived from Data File 2) as well as for B̂ 0
λ (Data File 5).
measured refractive increment of Hb solutions [11] and the water RI The uncertainties for the curve B̂
λ were computed and
[19] (measurement uncertainty indicated by shaded band) and different reveal that its shape is resolved much more accurately than
KK analyses. Concentration is c Hb  287 g L−1 . The method in Ref. [9] in the measurements B 
λ in Ref. [11]. The analysis, further-
ignores absorption outside the measured range and in Ref. [10] the UV more, shows that the real refractive increments for oxygenated
and IR absorption of water was taken into account. In the present work, and deoxygenated hemoglobin differ significantly from each
the excluded water fraction was accounted for, as well as a model for deep
other for wavelengths between 350 and 600 nm. In the vicinity
UV absorbance. All KK methods were applied to the same data set
depicted in Fig. 1. Curves calculated according to Refs. [9,10] are
of the absorption band at 420 nm, deviations between our result
matched to the experimental curve at λ0  800 nm. The two free for the real refractive increment and the experimental values [11]
parameters characterizing the deep UV model in our method are deter- clearly exceed the measurement uncertainties, where the curve
mined by a global fit to the experimental refractive increment. obtained by our KK analysis has a much smoother shape and
does not exhibit the nonmonotonic up-and-down movements for
wavelengths larger than 600 nm of the data presented in Ref. [11].

APPENDIX A: NUMERICAL INTEGRATION

SCHEME FOR KK RELATIONS
For numerical integration of KK relations, we follow the con-
cept described in Ref. [22]: a Riemann sum with Taylor expan-
sion at the singularities of the integrand. Numerical stability
was tested by comparison with the analytical transformations
of different Lorentzian and rectangular profiles.
We want to evaluate numerically the expression
Z λ
b λ λ
πG
λ  −2 α
ΛdΛ
λa Λ Λ 2
− λ2
Z λ

b 2 1 1
 − − α
ΛdΛ: (A1)
λa Λ Λλ Λ−λ
The measurement data are given at discrete points
Fig. 6. Residuals between calculated real RI n
λ and experimental αi ≔α
λi ; (A2)
real RI n
λ [11] for our method and the method applied in Ref. [10]

1
[cf. Fig. 5]. λi ≔λa  t i − ; i  1; …; N ; (A3)
2
where λ1  250, λN  1100, and t  1 nm. Hence
n
λ  nH2 O
λ  iκ H2 O
λ  c Hb B
λ  iα
λ; (35) λa  249.5, λb  1100.5 nm, and N  851. We only evalu-
ate the integral at the grid points G i ≔G
λi . The third term in
where κ H2 O
λ is negligibly small for λ ∈ 250; 1100 nm and the integral has a singularity at λ  Λ. The first two terms
physiological Hb concentrations c Hb . In the present work, we are not singular; hence, no principal value integrals have to
have computed the real refractive increment B
λ from exper- be used here. All integrals for nonsingular integrands are ap-
imental spectra of the imaginary refractive increment α
λ for proximated by Riemann sums, including the third term for
λ ∈ 250; 1100 nm. We formally separated the solution’s Λ∉λi − t∕2; λi  t∕2. The remaining principal value integral
imaginary RI into a water and a Hb part and then applied can be rewritten by Taylor series expansion of the integrand.
the KK relations to obtain the real RI and thus an expression Using only the lowest nonvanishing order yields
 Research Article Vol. 55, No. 31 / November 1 2016 / Applied Optics 8959

N

X  X
N
2 1 1 In Refs. [8,11], spectral reflectance measurements were per-
πG i ≈ − αj t − αj t − tαi0 : (A4) formed and evaluated using the Fresnel equation, Eq. (8), to
j1
λj λj  λi j≠i
λ j − λ i
j1 obtain the real RI n
λ. To determine the experimental value
of the real refractive increment B 
λ, several curves at different
Numerically, we use the nearest-neighbor lattice-derivatives,
8 concentrations c Hb were recorded, and the slope of n
λ with
<
αi1 − αi−1 ∕2 1 < i < N respect to c Hb was computed for all λ and normalized to the
tαi0  α2 − α1 i1 : (A5) known water RI to obtain β
λ.
:
αN − αN −1 iN Having this in mind, measurement uncertainties of the fol-
Note that Eq. (A4) can also be written as G  Kα, where K is a lowing quantities need to be taken into account:
N × N matrix. 1. Detector/instrument noise in the absorption spectra.
This affects the inverse absorption length μa
λ in Ref. [8]
APPENDIX B: UV DOMAIN KK TRANSFORM OF and the molar extinction coefficient ϵM
λ in Ref. [14].
LORENTZIAN CURVE 2. Detector/instrument noise in the reflectance spectra
[8,11], resulting in a wavelength-independent uncertainty of
If the KK transform (Eq. (10) is applied to the antisymmetric 3 × 10−6 L g−1 [11] for the real refractive increment β
λ.
Lorentzian line in the wavelength domain Eq. (14), the result is 3. Uncertainties of the solutions’ Hb concentration c Hb .

 4. Uncertainty of the Hb density ρHb relating mass con-
1 λ−L L
KαL 
λ  aL  centration c Hb and volume fraction ϕ. Here we assume one
π
λ − L2  Γ2 L2  Γ2

 digit, i.e., uρHb  10 g L−1  0.75%ρHb .
1 λL L 5. Uncertainty of the complex RI of water. This influence
− aL − : (B1)
π
λ  L2  Γ2 L2  Γ2 is negligible.
In our work, however, we need to integrate only over the deep These uncertainties can be divided into two classes, separat-
UV spectrum below a threshold λa . The analytical expression ing 1–2. from 3–4.:
for this reads for λ > λa. Equation (B2) is useful to describe the
(i) Noise in the measured reflectance and transmittance
deep UV part of the absorption spectrum without the need for spectra, affecting the spectral data locally. The corresponding
numerical integration quantities are labeled with the subscript “noise.” As a model,
Z

1 λa 1 1 we assume white noise (i.e., two measurement errors at differ-
G L
λ  − α
ΛdΛ ent wavelengths are not correlated).
π −λa Λ Λ − λ L

 (ii) Measurement uncertainties in scalar quantities
c Hb ; ρHb 
aL 1
λa − L2  Γ2 occurring as prefactors in the spectra. These affect the spectra
 2 ln
π 2
λa  L2  Γ2 by a global factor (i.e., the errors at different wavelengths have a
  correlation coefficient of 1 ). The corresponding quantities
Γ Γ 2Γ
×  − are labeled with a subscript “conc.”

λ − L2  Γ2
λ  L2  Γ2 L2  Γ2

    In addition, there are model errors, which arise due to un-
 λ −λ  Γ Γ
− ln  a  × − known absorption spectra outside the measured range and
λa  λ
λ − L2  Γ2
λ  L2  Γ2

 which we discuss in Subsection E of Appendix C.
Γ Γ
 π − arctan − arctan B. Error Model And Uncertainty Propagation
λa − L λa  L
 
For any random vector ξ ∈ RN and any nonrandom linear
λ−L λL 2L transform A ∈ RN ×N , we have for η  Aξ
× −  : (B2)

λ − L2  Γ2
λ  L2  Γ2 L2  Γ2
E
η  AE
ξ; (C1)
We also define the transformation for unit amplitude G̃ L
λ,
such that G L
λ  aL G̃ L
λ.
V
η  AV
ξAT ; (C2)
APPENDIX C: DETAILS OF UNCERTAINTY where E denotes the expectation value (or mean) and
PROPAGATION
V
ξij  cov
ξi ; ξj   E
ξi − E
ξi ξj − E
ξj ; (C3)
A. Uncertainties of Input Data
We now recapitulate how spectral data are measured, in order denotes the covariance matrix. This is independent of the
to understand which contributions to the combined uncertain- underlying probability distribution, only implying that the first
ties occur in the problem at hand. and second moments exist. We restrict our uncertainty analysis
The absorption spectra (inverse absorption length μa
λ, imagi- to mean values and covariance matrix, which describes the cor-
nary RI κ
λ) are measured via the attenuation of a beam of light responding probability distributions fully only in the special
passing through a sample of known thickness. This is performed case of a normal distribution.
for a number of wavelengths λi , i  1; …; N . The molar extinc- The uncertainties of the literature absorption spectra are not
tion coefficient ϵM
λ and the imaginary refractive increment α
λ given; neither the noise nor the concentration uncertainties are
are obtained from these quantities by dividing with the separately quantified. We thus estimate and model them as follows. The
determined molar/mass concentration of Hb in the solution. covariance matrix for the absorption spectra is assumed to be
 8960 Vol. 55, No. 31 / November 1 2016 / Applied Optics Research Article

V
α  Vnoise
α  Vconc
α. Here, This decomposes into a noise and a concentration term
Vnoise
α  σ 2noise;rel diag
α ;2
(C4) V
y  V
ynoise  V
yconc . For the contributions from con-
centration uncertainties, one finds [cf. Eqs. (C7) and (C12)]
is the local/uncorrelated part due to (white) detector noise. This
X
4
matrix is diagonal. V conc
y  vj vTj ; (C13)
Vconc
α  σ 2conc;rel ααT (C5) j1

is the global/correlated part due to concentration uncertainty. with

This matrix has a tensor-product structure. We use σ noise;rel  v1 ≔σ conc;rel B  ; v2 ≔σ conc;rel G VIS ;
0.5% for the relative noise level and σ conc;rel  1% for the rel- u Hb (C14)
v3 ≔σ conc;rel G UV ; v4 ≔ ρρHb G H2 O :
ative concentration uncertainty [24].
Equation (C5) is motivated as follows: Concentration errors Similarly, we obtain Vconc
B    v1 vT1 .
change a spectrum by a global factor, such that the measured However, for weighting the linear fit, we do not use this covari-
value ance matrix, but only the noise terms [i.e., V  Vnoise
y] with
αmeas 
1  ξα; (C6) Vnoise
y  Vnoise
B    KVnoise
αVIS KT  KVnoise
αUV KT ;
is off the (unknown) true value α. We ffi assume this error to be
pffiffiffiffiffiffiffiffiffiffiffi (C15)
unbiased (i.e., E
ξ  0 and var
ξ  σ conc;rel correspond to 
the relative concentration uncertainty of the solution). That is, (noise in nH2 O is negligible). Here Vnoise
B  is a diagonal matrix
E
αmeas   α, and containing the uncertainties given for the refractive increment in
Ref. [11]. The reason to use V  Vnoise
y is that the tensor-
V
αmeas   var
ξααT  var
ξ E
αmeas E
αmeas T product structure [cf. Eqs. (C7) and (C13)] of the systematic
|fflffl{zfflffl}
σ 2conc;rel covariance matrices makes them singular, and hence, the full
covariance matrix (including noise) is close to singular, which
≈ σ 2conc;rel αmeas αTmeas ; (C7) is a problem for the numerics. But after all, weighting with un-
which is Eq. (C5). In the following, we will drop the distinction certainties that arise from a global prefactor for all values makes
between unknown true values and measured values and substi- little sense.
tute the latter for the former. Since fˆ  Fy [Eq. (25)] is obtained by linear transforma-
This error model is formally invariant under linear transfor- tion, the covariance matrix is formally obtained as
mation. The measured value of some quantity obtained by a Vnoise
fˆ   FVFT  H
HT V−1 H−1 HT : (C16)
linear transform A (e.g., the KK transform)
For the concentration contributions one obtains
ηmeas  Aαmeas 
1  ξAα 
1  ξη; (C8)
X
4
is formally identical to Eq. (C6) such that the variance V
ηmeas  Vconc
fˆ   w j wTj with w j  F vj : (C17)
can be computed from E
ηmeas  in the same manner as for the j1
original quantity α, i.e., V
ηmeas   var
ξηηT .
Although fˆ is formally the result of the fit, the quantity
C. Uncertainty Propagation in KK Relations we are interested in is not fˆ , but rather,
For discrete data points, the KK relations can be written in ma-
trix vector form (i.e., G  Kα) where K is a N × N matrix (see B̂  B 0  fˆ  B 0  Fy  B 0  F
B  − B 0  
F − 1B 0  FB  ;
Appendix A). If V
α is the (co-)variance matrix of the spec- (C18)
trum α, then the covariance matrix of the transform G is
where 1 denotes the identity matrix. We can assume B  (the
V
G  KV
αKT ; (C9) measured refractive increment) and B 0 (the part of the refrac-
and with the model for V
α one obtains tive increment computed from absorption spectra and water
RI) to be uncorrelated, and we obtain
V
G  Vnoise
G  Vconc
G; (C10)
where Vconc
G  σ 2conc;rel GG T
has the same tensor product V
B̂ 
F − 1V
B 0 
FT − 1  V
B  FT
structure as Vconc
α in Eq. (C5). In contrast, Vnoise
G   V
B 0   V
fˆ  − FV
B 0  − V
B 0 FT : (C19)
KVnoise
αKT is, unlike Vnoise
α, not diagonal, because of
the nonlocality of the KK transform. This also decomposes into

D. Uncertainty Propagation in Linear Fit V
B̂  Vnoise
B̂  Vconc
B̂; (C20)
The covariance matrix V
y of where both parts are computed separately. Vconc
B̂ contributes
y  B  − B 0  B  − G VIS − G UV − G H2 O ; (C11) more strongly to the diagonal elements of V
B̂ (i.e., the var-
iances of the B̂ i ) than Vnoise
B̂. On the other hand, Vconc
B̂,
is obtained by is strongly correlated among all elements, whereas Vnoise
B̂ is
V
y  V
B    V
B 0  not. To illustrate the degree of correlation, the numerical
derivative B 0
λi , obtained from finite differences, is plotted
 V
B    V
G VIS   V
G UV   V
G H2 O : (C12) in Fig. 4 with the corresponding standard deviations.
 Research Article Vol. 55, No. 31 / November 1 2016 / Applied Optics 8961

D.1. Deoxyhemoglobin 4. E. Schonbrun, R. Malka, G. Di Caprio, D. Schaak, and J. M. Higgins,

The refractive increment for deoxyhemoglobin in Eq. (26) can “Quantitative absorption cytometry for measuring red blood cell hemo-
globin mass and volume,” Cytometry A 85, 332–338 (2014).
also be written as 5. Y. Kim, H. Shim, K. Kim, H. Park, S. Jang, and Y. Park, “Profiling indi-
deoxy
B deoxy  G lit − FG lit −
F − 1G H2 O  FB  ; (C21) vidual human red blood cells using common-path diffraction optical
tomography,” Sci. Rep. 4, 6659 (2014).
from which follows 6. D. H. Tycko, M. H. Metz, E. A. Epstein, and A. Grinbaum, “Flow-cytometric
light scattering measurement of red blood cell volume and hemoglobin
deoxy
Vnoise
B deoxy   Vnoise
G lit  concentration,” Appl. Opt. 24, 1355–1365 (1985).
7. N. Mohandas, Y. R. Kim, D. H. Tycko, J. Orlik, J. Wyatt, and W.
 FVnoise
G lit FT  FVnoise
B  FT ; (C22) Groner, “Accurate and independent measurement of volume and he-
moglobin concentration of individual red-cells by laser-light scatter-
X
6
ing,” Blood 68, 506–513 (1986).
Vconc
B deoxy   uj uTj ; (C23) 8. M. Friebel and M. Meinke, “Determination of the complex refractive
j1 index of highly concentrated hemoglobin solutions using transmit-
tance and reflectance measurements,” J. Biomed. Opt. 10, 064019
with (2005).
uj ≔Fvj for j  1; 2; 3; u4 ≔
F − 1v4 9. D. J. Faber, M. C. G. Aalders, E. G. Mik, B. A. Hooper, M. J. C. van
deoxy deoxy (C24) Gemert, and T. G. van Leeuwen, “Oxygen saturation-dependent
u5 ≔σ conc;rel G VIS ; u6 ≔σ conc;rel G UV : absorption and scattering of blood,” Phys. Rev. Lett. 93, 028102
(2004).
E. Peptide Absorption: Influence of Peak Shape 10. O. Sydoruk, O. Zhernovaya, V. Tuchin, and A. Douplik, “Refractive
We have assumed the exact shape of the peptide peak to be of index of solutions of human hemoglobin from the near-infrared to
the ultraviolet range: Kramers–Kronig analysis,” J. Biomed. Opt.
minor importance, such that the results depend mainly on its 17, 115002 (2012).
position, total strength, and width. In order to justify this 11. M. Friebel and M. Meinke, “Model function to calculate the refractive
assumption quantitatively, we evaluated an alternative model, index of native hemoglobin in the wavelength range of 250–1100 nm
where the peak has a rectangular shape. dependent on concentration,” Appl. Opt. 45, 2838–2842 (2006).
The comparison between results with the Lorentzian peak 12. H. J. Deuling and W. Helfrich, “Red blood cell shapes as explained on
the basis of curvature elasticity,” Biophys. J. 16, 861–868 (1976).
centered at L  187 nm and half-width at half-maximum Γ  13. V. Heinrich, K. Ritchie, N. Mohandas, and E. Evans, “Elastic thickness
11.6 nm and a rectangle centered around LΠ  L and half- compressibility of the red cell membrane,” Biophys. J. 81, 1452–1463
width ΓΠ  Γ reveals that deviations between the two models (2001).
are negligible in comparison to the propagated data uncertain- 14. Y. Sugita, M. Nagai, and Y. Yoneyama, “Circular dichroism of hemo-
ties for the majority of wavelengths. Only at wavelengths λ < globin in relation to the structure surrounding the Heme,” J. Biol.
Chem. 246, 383–388 (1971).
400 nm does the deviation jB̂
λ − B̂ Π
λj exceed the esti- 15. S. Prahl, “Optical absorption of hemoglobin,” 1999, http://omlc.org/
mated uncertainties due to noise, but is still smaller than spectra/hemoglobin/.
the total uncertainty, including concentration errors. For 16. D. J. Segelstein, “The complex refractive index of water,” Ph.D. dis-
λ > 400 nm, the deviation is at least 1 order of magnitude sertation (University of Missouri, 1981).
17. G. Braunitzer, “The molecular weight of human haemoglobin,”
smaller than the total uncertainties.
Bibliotheca Haematologica 18, 59–60 (1964).
18. H. Arwin, “Optical properties of thin layers of bovine serum albumin,
Acknowledgment. The authors would like to thank γ-globulin, and hemoglobin,” Appl. Spectrosc. 40, 313–318 (1986).
Moritz Friebel for providing in tabulated form the absorbance 19. M. Daimon and A. Masumura, “Measurement of the refractive index of
data for oxygenated and deoxygenated hemoglobin published distilled water from the near-infrared region to the ultraviolet region,”
Appl. Opt. 46, 3811–3820 (2007).
in Ref. [8] as figures.
20. A. R. Goldfarb, L. J. Saidel, and E. Mosovich, “The ultraviolet absorp-
tion spectra of proteins,” J. Biol. Chem. 193, 397–404 (1951).
21. A. H. Woods and P. R. O’Bar, “Absorption of proteins and peptides in
REFERENCES the far ultraviolet,” Science 167, 179–181 (1970).
1. R. Barer and S. Joseph, “Refractometry of living cells: Part I. basic 22. C. A. Emeis, L. J. Oosterhoff, and G. de Vries, “Numerical evaluation of
principles,” Q. J. Microsc. Sci. s3-95, 399–423 (1954). Kramers–Kronig relations,” Proc. R. Soc. London 297, 54–65 (1967).
2. Y. Park, T. Yamauchi, W. Choi, R. Dasari, and M. S. Feld, 23. L. H. Laasberg and J. Hedley-Whyte, “Optical rotatory dispersion of
“Spectroscopic phase microscopy for quantifying hemoglobin concen- hemoglobin and polypeptides: effect of halothane,” J. Biol. Chem.
trations in intact red blood cells,” Opt. Lett. 34, 3668–3670 (2009). 246, 4886–4893 (1971).
3. Y. Sung, N. Lue, B. Hamza, J. Martel, D. Irimia, R. R. Dasari, W. Choi, 24. K. Witt, H. U. Wolf, C. Heuck, M. Kammel, A. Kummrow, and J.
Z. Yaqoob, and P. So, “Three-dimensional holographic refractive- Neukammer, “Establishing traceability of photometric absorbance val-
index measurement of continuously flowing cells in a microfluidic ues for accurate measurements of the haemoglobin concentration in
channel,” Phys. Rev. Appl. 1, 014002 (2014). blood,” Metrologia 50, 539–548 (2013).