METHOD OF MEASURING TISSUE HEMOGLOBIN SATURATION USING GAUSSIAN DECOMPOSITION
Field of the Invention
The present invention relates to determining hemoglobin saturation
in tissue, and, in particular, to a method of using gaussian decomposition
of diffuse reflectance spectra to determine hemoglobin saturation in
tissue.
Background of the Invention
The principal function of hemoglobin is to transport oxygen from
the lungs to body tissues. If the transport process is temporarily
interrupted at any step between the binding of oxygen to hemoglobin in
the lungs and its final conversion to water in the cells of the tissue, the
result will be a decrease in intracellular adenosine triphosphate, a
molecule used to store energy for cellular processes. If the interruption
continues over a prolonged time interval, the cells' inability to perform
routine cellular functions will eventually result in cell death.
Numerous pathologic mechanisms can reduce the amount of
oxygen in tissues (tissue hypoxia). Moreover, a number of surgical
procedures, designed to repair or alter parts of a body's circulation
system, employ techniques that intentionally interrupt oxygen transport.
In the 1970s, Ohmeda introduced the pulse oximeter. This is an
inexpensive instrument that can be routinely applied to patients in a
variety of settings; at the patients bedside, in the operating room, and has
even become standard equipment on emergency vehicles. The device
allowed medically trained personnel, for the first time, to monitor a
patient's arterial hemoglobin saturation, without having to place an
arterial line. The instrument has achieved wide success and has become
entrenched as a standard for medical care. However, arterial hemoglobin
saturation is a parameter that conveys to the physician how well the heart
and lungs are functioning in their ability to keep blood oxygenated. It
does not provide any information about how well the tissues that use the
oxygen are performing, nor does it provide information about their status.
It simply provides the physician with a number that relates to the amount
of oxygen that is available to tissues, provided the tissues are functioning
normally and have available to them a means of drawing on this oxygen
pool.
In instances where blood pressure is low as a result of hemorrhage,
blood may be adequately oxygenated, but unavailable to tissues because
of a low perfusion pressure, or because the body has restricted the oxygen
supply to that tissue in an attempt to conserve other organs, such as the
heart and brain. Another instance where arterial hemoglobin saturation is
of no value is in the case of endartarectomy. In this common medical
procedure designed to physically remove placque from the inner surfaces
of blood vessels supplying the brain, a surgeon may choose to apply a
clip across the affected side as he performs the procedure. He does this to
prevent blood loss and experience has shown that the clip can remain in
place for a finite period of time before it has to be removed to prevent
damage to the brain. The amount of oxygen that the brain on the operated
side receives during the procedure is dependent upon the degree to which
blood vessels cross-connected to the unoperated side can supply the
affected side. The amount of "collateralization" is dependent upon
numerous factors. Among these are the age of the patient, the size and
number of vessels, and the degree and stage of disease. Presently, the
surgeon relies upon indirect measures of stress to determine when and if
the brain's oxygen supply is compromised. These indirect measures are
other technologies that monitor aspects of function, such as evoked
potentials, or electroencephalographic activity. But, these measures
"report" deficiency only after function is compromised. Presently, there
are no available technologies in common clinical use that are capable of
forewarning the physician or surgeon that a state of oxygen deficiency is
either developing, or exists, in a specific organ's tissues.
A number of 'laboratories are presently working to develop "tissue
oximeters". Early efforts focused upon oximeters directed at monitoring
the degree of saturation of hemoglobin in brain tissue using optical
methods that that can be performed noninvasively. Focus has been on
brain tissue, not only because of the importance of this organ, but also
because of the apparent advantages offered by the anatomy of the head.
The brain is relatively near the surface of the skin and is separated from
the environment by only a thin layer of optically transparent bone that
contains a minimal amount of blood. Measuring brain tissue hemoglobin
saturation has turned out to be a not so easy task to perform.
At least six different methodologies have been developed and used
to extract hemoglobin related information from brain tissue spectra. The
goal of each of these approaches is to quantify the two components of
hemoglobin individually, and in such a manner, as to allow the
components to be recombined in the form of a ratio that describes the
percentage of hemoglobin, contained within an optical field, that is
combined with oxygen. Generally, the methods that have emerged to
perform this task have done so from new insights or from new theoretical
prospectives that offer a potential solution to the tissue hemoglobin
saturation problem. Accordingly, instrumentation has been tailored to the
specific requirements of the analytical solution conceived.
Simple, multi-wavelength, absorbencv-based algorithms
The distinguishing characteristic of approaches under this category
is that only optical attenuation is used in the computations required to
obtain a numeric solution. Typically, this information is input into an
algorithm in the form of optical density measurements monitored at two
to four select wavelengths. Historically, in vivo multi- wavelength
spectroscopy was first employed as a means of characterizing the in vivo
redox state of cytochrome oxidase. To do this, it was recognized that the
various forms of hemoglobin had to be characterized as well, because
each overlies the cytochrome oxidase spectrum. While it remains a goal
of most in vivo spectroscopists to develop a method that fully
characterizes cytochrome oxidase, most of the recent efforts have focused
upon ways to fully characterize hemoglobin concentration change. Multi-
wavelength absorbency based spectroscopy is attractive because it
requires simple and straight-forward mathematical computations.
Instrumentation design is relatively uncomplicated and economically
feasible. A weakness of multi-wavelength absorbency-based methods is
that they employ no means of characterizing light lost from the sampled
tissue by light scattering processes, and thus, in principle, can only report
changes in constituent concentration from an arbitrarily established
baseline.
A number of instruments marketed today employ simple, multi-
wavelength, absorbency technology. These include products by the
Somanetics Corporation (INVOS 3100 and 4100), the Hamamatsu
Corporation (NIRO 500 and 1000), and NIM, Incorporated (Runman).
Two other instruments recently appeared in reports from the literature,
but there is little public information regarding the nature of their
technologies or methods for analytical solution. These are instruments
being developed by Hutchinson Technologies, Inc. of Hutchinson, MN
and by the Tostec Co., Inc. of Tokyo, Japan. All of these instruments
claim to be able to provide trend information that is based upon an initial
baseline setting and all claim to provide an accurate measure of tissue
hemoglobin saturation. Because the Somanetics INVOS 3100 instrument
was the first commercially available instrument in this country, it has
undergone the most evaluation. These studies have been disappointing in
some respects, for the clinician's confidence in this potentially useful
technology has waned in the aftermath. The INVOS 3100 instrument has
been revised and new reports on the INVOS 4100 are beginning to
appear.
Derivative spectroscopy
If it can be assumed that light scattering primarily influences
baseline absorbency, i.e., that light scattering is not strongly wavelength
dependent, then baseline influences can be nulled by derivatizing spectra.
In 1989, it was shown that it is possible to compute a parameter, using
derivatized spectra, that correlates highly with brain tissue hemoglobin
saturation. While the method was demonstrated to be feasible, it was at
that time believed to be impractical for commercialization because the
algorithm was not portable to other instruments, i.e., the coefficients
employed were instrument specific.
Pseudo-random modulated code spectroscopy (PRM)
PRM technology was originally developed for use in satellite
technologies to measure the height of the ocean surface from space.
PRM spectroscopy evolved as a means of quantifying the much shorter
distances that photons travel in brain tissue. The PRM technique relies
upon a randomized sequence of light pulses of varying, but known,
widths. These pulse sequences are unique and when repetitively
transmitted into tissue and their emergence from the tissue is measured,
the sequences can be identified using autocorrelation analysis. By
comparing the input and output sequences, a time-shifted difference, or
time-of-travel for photons through tissue, can be characterized. The
scattering and absoφtion coefficients (μs and μa, respectively) can be
extracted from X2 curve fitting the convolution of the instrument impulse
response function with the theoretical reflectance function of μs and μa ,
respectively to the measured photon migration profile. Blood
oxygenation can be measured by performing the calculations at two
wavelengths and ratioing the absoφtion coefficients found there. The
advantages of the PRM technique is that it can be performed using
instrumentation that is relatively inexpensive to manufacture and safe to
use.
Time-resolved spectroscopy (TRS)
Because tissue is an effective multiple scatterer of light, the light
pathlength needed to perform quantitative calculations of hemoglobin
concentration is normally unknown. In 1988, it was shown that it was
possible to calculate a mean pathlength by measuring the time it takes
energy, pulsed in picosecond increments, to traverse the head. Time-
resolved spectroscopy has become a standard in the field of in vivo
spectroscopy because of its ability to measure pathlength directly.
However, it has not been demonstrated to be a method suitable for
clinical use for several reasons. TRS requires using very short laser
pulses and ultrafast photon counting equipment. Consequently, the
delivery of light and the recovery of light from tissues requires equipment
that is expensive, bulky, and requires a high degree of maintenance.
Moreover, the intrinsic high peak power of picosecond lasers requires
further operational precautions.
Phase-modulation, or frequency-resolved spectroscopy (FRS)
FRS followed as a natural application of previous experiences with
radar and altimeter development. A coded light signal (typically
sinusiodal or square in shape) is injected into tissue. Photon diffusion
encodes the tissue characteristics in the timing of the delayed, received
pulse and in the intensity of the time profile. Thus, instead of receiving
a clean replicate of the transmitted pulse, the returned signals are spread
out over time and are diminished in amplitude. The signal information is
decoded using procedures similar to that described previously for the
PRM method. This technique, like that of PRM, requires relatively
inexpensive equipment to generate encoded light signals. It is safe to use
in that only small amounts of energy are required (microwatts). It
provides a measure of pathlength, thus allowing constituent concentration
to be computed directly from the Beer-Lambert relationship. The
disadvantage of the method is that it may be more susceptible to noise
than is PRM. Also, like PRM methods, FRS relies upon a mathematical
model of photon migration in a semi-infinite medium to recover tissue
absoφtion coefficients.
Two instrument designs have been based on the FRS methodology.
The NIM instrument (PMD 4002, MM Incoφorated, Philadelphia, PA)
incoφorates three tissue-illuminating wavelengths, and a fourth,
reference wavelength. It is a hybridized design, thus, incoφorating both
FRS and multi- wavelength, absorbance-based methodologies. It has
been tested, both in vitro, and in vivo, in animal experiments. Tissue
hemoglobin saturation measurement made with this instrument system
show an excellent linear correlation with tissue hemoglobin saturation
estimated from measurements made on whole blood. Another instrument
is presently marketed by ISS, Inc. This instrument has not undergone
rigorous evaluation. It is a two-wavelength, four-channel device.
Photoacoustic spectroscopy
When light energy is absorbed by molecules of a tissue, heat is
liberated in the process. This quickly results in thermal expansion of the
surrounding tissue, which, in turn, is followed by a rapid collapse to the
original state. The result is an acoustic signal that can be "heard" in the
ultrasound range. The amplitude of the acoustic signal is proportional to
the amount of light absorbed. Most importantly, the amplitude of the
signal is related to the total amount of light absorbed, including that
which is contained in the photon migration path. It has been
demonstrated that this method is feasible when applied to biologic tissue.
However, while the method is advantageous in being insensitive to
changes in the light scattering properties of tissue, it was later found that
the useable signal originates close to the surface of the tissue, and was,
thus, not suited for measuring changes in tissue constituent composition
in deep structures.
At least six different approaches have been used to extract
constituent information from spectra collected from tissues. All of the
techniques described use procedures that are noninvasive and non-
damaging to the tissue. With the exception of TRS, most of these
methods can be performed using equipment that is relatively inexpensive
to construct and maintain. TRS, FRS, and PRM technologies are at the
forefront of development because they incoφorate solutions for
measuring pathlength, a necessary component of the Beer-Lambert
solution for absolute constituent concentration. Multi-wavelength
methodologies have become established for trending measurements, but
they cannot be used for quantitative measurement. All methods are
susceptible to noise because of the low light level requirements
established for patient safety. Excepting derivative spectroscopy, all
methods are multi-wavelength and use absorbency changes occurring at
only two to four points within the entire near infrared ("NIR") range.
Hence, all methods, to a greater or lesser extent, are subject to how well
absorbency changes at these wavelengths reflect changes occurring
within the monitored tissue.
Summary of the Invention
An object of the present invention is to provide a method of
measuring tissue oxygenation that can be performed using
instrumentation that is portable and inexpensive to make, such that
bedside and field measurements of this parameter can be quickly and
accurately made.
The present invention is a method for decomposing compound
diffuse reflectance spectra, collected from tissue, such as brain tissue, into
discrete constituents that can then be used to compute tissue hemoglobin
saturation. The present invention requires a spectrometer capable of
collecting full absoφtion spectra in the NIR band and a computer for data
acquisition and processing. Because constituents other than hemoglobin
must be quantified in the process of extracting this information, the
method additionally yields qualitative information about these
constituents. In the case of brain tissue, these secondary parameters are
related to the water content of the brain tissue, the redox state of brain
tissue cytochrome oxidase, cerebral tissue lipid content, and the brain's
light scattering properties. The secondary constituent characterizations
are qualitative, but the values can be used for trending puφoses, or for
comparing relative amounts between tissue regions or between brain
hemispheres, when collected and analyzed as a function of optical path
length. The method of the present invention can be further developed to
quantify all of these secondary parameters.
The constituents of cerebral tissues that contribute to light
absorbency, i.e., oxyhemoglobin, deoxyhemoglobin, water, lipid,
cytochrome oxidase and a component for characterizing light loss due to
scattering, are further characterized and used to construct a model system
that emulates cerebral tissue reflectance spectra in a variety of conditions.
Using this model system in a reverse mode, compound spectra collected
from brain tissue are decomposed into individual spectral features. The
values for features attributable to oxyhemoglobin and deoxyhemoglobin
are then used to construct a ratio that quantifies the percentage of total
hemoglobin that contains oxygen. Because the major portion of light
collected by the detecting element of the equipment has transited through
brain tissue, this ratio becomes a quantitative measure of brain tissue
hemoglobin saturation. Because it is a ratio, it does not require a
knowledge of in vivo molar absoφtion coefficients and it is relatively
pathlength insensitive .
BRIEF DESCRIPTION OF THE DRAWINGS
FIGURE 1 is a graph showing the absoφtion spectra of
oxygenated and deoxygenated hemoglobin.
FIGURE 2 is a block diagram of the instrumentation used to carry
out the method of the present invention.
FIGURE 3 is a graph showing NIR prediction versus measured
tissue hemoglobin saturation.
FIGURE 4 is a graph showing that the oxyhemoglobin feature
increases during hypoxic-hypoxia.
FIGURE 5 is a graph showing that adding linkage between the 760
nm and 930 nm features corrects the directional error in recovered
oxyhemoglobin
FIGURE 6 is a graph showing the ratio of oxyhemoglobin
absoφtion to absoφtion attributable to total hemoglobin is proportional
to tissue hemoglobin saturation.
FIGURE 7 is a graph showing that the ratio of oxyhemoglobin
absoφtion to absoφtion attributable to total hemoglobin can be adjusted
using two coefficients to provide a numerical predictor of brain tissue
hemoglobin saturation.
FIGURE 8 is a graph showing the best and worst case fits to
absoφtion spectra and the error of prediction.
FIGURE 9 is a graph showing that feature absorbency is optrode
separation distance dependent.
FIGURE 10 is a graph showing tissue hemoglobin saturation,
measured by gaussian decomposition methods is optode separation
distance independent.
FIGURE 11 is a graph showing water corrected spectra from bacon
and white matter.
FIGURE 12 is a graph showing the result of fitting the HbSatbt
ratio to computed tissue hemoglobin saturation.
FIGURES 13 A and B are graphs showing the effect linking two
deoxyhemoglobin features has on the total hemoglobin attenuation.
FIGURE 14 is a graph showing an absoφtion spectrum from cat
head and the goodness-of-fit provided by Model C.
FIGURE 15 is a graph showing an absoφtion spectrum from
gastrocnemius and the goodness-of-fit provided by Model C.
DESCRIPTION OF THE PREFERRED EMBODIMENT
An absoφtion spectrum is a plot of how light within a range of
energies is absorbed by molecules. For pure substances (constituents) in
a dilute solution, the amount of light absorbed at any particular energy
level is proportional to the concentration of the constituent in the solution.
The spectrum for a particular constituent can be thought of as being
analogous to a fingeφrint, with it's shape being it's most distinguishing
characteristic. It may be simple or complex, depending on the number of
absorbers present, but it is uniquely shaped. For mixtures of constituents
in a solution, a compound spectrum is observed. This compound
spectrum results from the summation of all absorbencies by all of the
substances contained within the mixture. Compound spectra can be
broken down into basic elements if the features of all of the possible
constituents are known.
The Beer-Lambert law states that absoφtion, A, at a selected
wavelength, λ, is proportional to the concentration of a constituent, c, and
the pathlength, L, through which the light travels during measurement, so
that:
A(λ) = ε(λ)cL (1)
where ε(λ) is the molar absoφtion coefficient and is wavelength
dependent. For many absorbers, ε(λ) has the characteristics of a gaussian
probability distribution and is the property that lends shape to the
absoφtion spectrum. For simple constituents (constituents with a single
absorbing center) having these properties, an equation of the form
[Aj = εmxi c L EXP((- .5((λ-λc)/FWHM) 2) (2)
may be written, where [A]χmin..λmaχ s an array, or range of absorbencies,
or an absoφtion spectrum, εmax is the maximum value of the absoφtion
spectrum within the range, FWHM is the width of the spectrum at
l/2*εmax, and λc is the wavelength around which the spectral range is
centered. Complex spectra, i.e. constituents with multiple absorbing
centers, can be described as a summed representation of equation 2,
wherein each center is described. Equation 2, or its summed
representation, thus describes constituent shape; a property that is unique
for every substance. Note that when λc equals λ, the terms inside the
brackets reduce to zero and the exponentiated portion of the equation
takes on the value of unity. In this special case, equation 2 reduces to
equation 1. It follows that c may be determined from [A]λmin..χmax, if εmax,
L, FWHM, and λc are known.
Decomposition of compound spectra requires an a priori
knowledge of the absoφtion properties of all constituents that absorb
within the specified region of optical monitoring. This information may
be termed a spectral feature database. Because some compounds have
similar features at similar energy locations, rules regarding how features
interact must also be known before they can be separated properly. For
example, the deoxyhemoglobin molecule has two absorbencies in the
700-1100 nm NIR band, one of which overlaps with oxyhemoglobin. To
separate a compound spectrum containing oxyhemoglobin and
deoxyhemoglobin into elemental spectra, a rule is needed indicating that
the deoxyhemoglobin contribution to the region overlapping
oxyhemoglobin will be a constant fraction of the non-overlapping feature.
One other factor must be known. Because photons shone into tissues are
also subject to light scattering, the knowledge base must also include
rules to account for "apparent" absoφtion, or light loss. The spectra
feature database, used in conjunction with a rule set and a definition
regarding how light scattering contributes to baseline absoφtion, is herein
defined as a knowledge base or, in engineering terminology, a model
system. It follows that, if the model system completely describes the real
system, then it can be applied to real spectra in a reverse manner and used
to derive the relative concentrations of the constituents that contribute to
a compound spectrum. A least squares curve fitting procedure is used to
adjust the parameters of a model system until the error between the
experimental spectrum and the model spectrum is minimized. The
individual contributing constituent information can then be recovered.
The constituents of cerebral tissues that contribute to light
absorbency are known. According to the present invention, these
constituents are further characterized and are used to construct a
constrained parameter, model system that emulates cerebral tissue
reflectance spectra in a variety of conditions. The knowledge base
incoφorates information about oxyhemoglobin, deoxyhemoglobin, water,
lipid, cytochrome oxidase and includes a component to characterize light
loss due to scattering. Using this model system in a reverse mode,
compound spectra collected from brain tissue are decomposed into
individual spectral features. The values for features attributable to
oxyhemoglobin and deoxyhemoglobin are "recovered" and used to
construct a ratio that quantifies the percentage of total hemoglobin that
contains oxygen. Because the major portion of light has transited through
brain tissue, this ratio becomes a quantitative measure of brain tissue
hemoglobin saturation. This has been experimentally verified. The
model system was built from spectral information collected in dogs and
cats, but is generally applicable to other mammalian species, including
man.
One hemoglobin molecule consists of two alpha and two beta heme
containing polypeptide chains (globins). Together, the chains form a
single molecule with four heme units that is capable of carrying four
oxygen molecules. The features present in a sample spectrum of
hemoglobin depend on the relative proportion of the number of
hemoglobin molecules that contain bound oxygen. In the fully
oxygenated form, features of high absorbency are located at 415, 542,
577, and 930 nm. In the deoxygenated form, the features shift location
and are located at 431, 555, 760, and 910 nm. Prior work in monitoring
brain tissue hemoglobin saturation has focused on identifying the
constituents in brain that have absorbing properties in a narrow energy
band known as the near-infrared window ("NIR"), i.e., 700 nm to 1100
nm. This is because this energy window contains only weak absorbencies
which allows light to penetrate more deeply into tissue than in other parts
of the light energy band. The features present in the NIR band, for
oxygenated and deoxygenated hemoglobin, are illustrated in Figure 1.
Above the upper energy cutoff of the NIR window, i.e., λ < 700 nm, a
strong absorbency by the visible hemoglobin features prevents light from
penetrating deeply. Below the lower energy limit, i.e., λ > 1100 nm, light
is absorbed superficially by a strong water absorbency. At least
identifiable features are contained within the NIR window of biological
tissues. Three features are attributed to hemoglobin, two to lipid, two to
cytochrome oxidase, and three to water. Of the four primary NIR
absorbers, water, hemoglobin, cytochrome oxidase, and lipid, water is the
most abundant absorber (82% by weight), followed by lipid (-12%),
hemoglobin (-0.4%), and cytochrome oxidase (<0.1%).
FIGURE 1 shows the absoφtion spectra 11 and 12 of oxygenated
and deoxygenated hemoglobin, respectively. The spectra 11 and 12 were
collected using a fourier transform spectrometer (Model 2000, Perkin-
Elmer Coφoration). The spectra 11 and 12 were collected in
transmittance mode while illuminating a sample of whole blood perfusing
through a 0.1 cm pathlength quartz flow-through perfusion cell that had
been inserted into an arterial-venous shunt, created in a cat animal model.
Hemoglobin saturation during the period when the oxygenated spectrum
was collected, measured 98.7% on a benchtop blood gas analyzer (ABL,
Radiometer). During collection of the deoxygenated spectrum 12, it
measured 25.3%. Both spectra have been post-processed and corrected
for instrument tilt and water absoφtion.
The absorbing properties of hemoglobin have been extensively
studied. In normal individuals, the predominant hemoglobin forms are
oxyhemoglobin and deoxyhemoglobin, although other forms of the
molecule may rise to significant concentrations in unhealthy individuals.
In the fully oxygenated state, oxyhemoglobin absorbs light in a broad
portion of the NIR band and has a peak absorbency centered at
approximately 930 nm. As oxyhemoglobin deoxygenates, absorbency at
930 nm decreases in magnitude and appears to shift to a new center
wavelength at 910 nm. Simultaneously, absorbency at all wavelengths
below approximately 815 nm increases and a new feature is seen to form
at 760 nm. It is commonly assumed that the changes in peak magnitude
at 930 nm and at 760 nm can be proportionately graded between the fully
oxygenated and fully deoxygenated states. Moreover, when the NIR
band features are used, it is common practice to ignore the visible and
ultraviolet band features.
The spectral characteristics of cerebral lipids have not been
published. Prior work shows that animal fat absorbs significantly at 928
nm and -1030 nm.
Table 1 itemizes the major NIR band absorbing constituents of
cerebral tissue and their center wavelength properties.
Table 1. Constituents of brain spectra published in the current literature.
Using data obtained from the literature, the individually
identifiable constituents of the NIR band have been characterized in a
manner that allows a knowledge base to be constructed. The knowledge
base consists of a series of equations that characterize each of the
absorbing features by their center wavelength, λc, and full width at half
maximum ("FWHM"). The shape of an absorbing feature is assumed to
have gaussian properties and is described as,
[OD]λmn λmax = hcEX?(-0.5((λ-λc)/FWHM) 2) (3)
where: [OOJλmm χmaκ is a range of optical density, or absorbence, and is
dimensionless; λ is wavelength, in nanometers; FWHM is the full width
at half-maximum, in nanometers; and hcis peak height; and λc it the peak
center wavelength, in nanometers. Note that here hc is a composite
parameter equal to εmΑxcL identified in equation 2, and is, therefore,
without dimensions. Thus, a compound tissue spectrum becomes the
summation of the simple constituent features superimposed on an
apparent baseline absoφtion attributed to light loss due to photon
scattering. Equation 3 is replicated n times, for the n absorbency peaks
occurring in the NIR window and a constant, k, is added. Thus, for a
compound or complex spectrum, shape is defined as
n
[OD]λτmn λmax = ∑ (ODλ)n + k (4)
which describes the general case knowledge base. Baseline OD is
assumed to be wavelength independent. To account for features outside
of the 700-1100 nm band that overlap into it and contribute absoφtion, it
was assumed that features at 555 nm and 577 nm "splay" into the NIR
band as a result of the high absoφtive properties. A first model system,
termed herein as Model A for descriptive puφoses, was then devised on
the basis of the features shown in Table 2.
Table 2. Features of the first model system used to extract tissue hemoglobin saturation from in vivo collected head spectra.
Animal Studies and Surgical Procedures
All experiments described herein were pre-approved by the Animal
Care and Use Committee of the Johns Hopkins University Medical
Center. In 21 dogs, anesthesia was induced using a combination of
sodium pentobarbital and fentanyl anesthesias, administered
intravenously. Anesthesia was maintained using only fentanyl. Cannula
were placed in the femoral arteries and veins to enable measurements of
arterial blood pressure and to collect arterial blood samples. A small
sized cannula was placed in the superior saggital sinus to collect samples
of cerebral venous blood. All blood gases were measured using a
standard laboratory blood gas analyzer (ABL, Radiometer Coφoration,
Copenhagen, DK). Hemoglobin saturation and content were measured on
an OSM3 (Radiometer coφoration, Copenhagen, DK). Brain tissue
spectra were collected through exposed skull after the temporalis muscle
was removed, by placing two fiber optic bundles directly on the skull
surface. The "optodes", that is, the fiber bundles delivering and
collecting light at the skull's surface, were sealed using light opaque black
modeling clay.
Several protocols for data collection were used, depending upon
the type of information sought.
Hvpoxic-hypoxia: All animals were challenged with multiple
levels of reduced inspired oxygen concentration. The order of
administering the hypoxic gas mixtures was usually from normoxic levels
to severely hypoxic levels, usually in five to eight decremental steps. At
each level, steady-state conditions were a requirement before collecting
paired arterial and cerebral venous blood samples for analysis. A sample
spectrum was collected either before or after the blood samples.
Hypercapnia: In twelve of these animals, C02 was separately
administered to induce tissue hyperoxygenation.
Anemic-hypoxia: In another two of the animals, anemic-hypoxia
was induced by infusing two liters of lactated ringers while exchanging
equivalent volumes of blood.
Hypothermia: In one animal, body temperature was cooled to 18
°C. Blood samples and spectra were collected during the period of body
temperature reduction and after cooling was achieved graded hypoxia
was induced stepwise to cover a large range of arterial blood hemoglobin
saturations.
Figure 2 is a schematic illustration of the instrumentation 10 used
with the above studies, and which could be used with the present
invention. Light is generated via a high intensity halogen lamp 13
powered by a power supply 14. The light 15 from lamp 13 is shined
through a lens 16 into a Perkin Elmer FTIR 2000, Fourier transform
spectrometer 18 (Model 2000, Perkin-Elmer Coφ, Norwalk, CT) that was
specially modified for this application by the instrument division of
Perkin Elmer. The modifications allowed for the use of external light
source 13 (Model 66183 halogen external light source; Oriel Instruments,
Stratford, CT) and for the instrument to pass it's light output 15 to the
input 20 of a 6-foot, 5 mm diameter, steel shielded fiber optic bundle 22,
also known as an optode. The output end 24 of bundle 22 was placed in
contact with the exposed skull 29 of an animal. Light 15 passing through
the bundle was deposited on the skulls surface 32, whereby some photons
migrated through the tissues of the skull, dura, cerebrospinal fluid, and
brain 30, to exit at a distant site on the skulls' surface 28. A portion of the
exiting light 26 was collected by a short, 3 -foot length of shielded fiber
optic bundle 34, also an optode, and passed to a peltier-cooled avalanche
photodiode 36 ("APD"); (Advanced Photonix, Inc., Camarillo, CA).
APD 36 is powered by a high voltage power supply 38. The signal from
APD 36 was then passed to the control of a computer 40 after being
filtered digitally by filter 42 (Model SR650, Dual channel,
Highpass/lowpass Programmable Filter, Stanford Research Systems, Inc.,
Sunnyvale, CA). The data acquisition and timing schemes for light
generation and collection were software controlled. Additional software
options allowed for control of the number of spectra collected, the
number of scans averaged, and output aperture control using neutral
density filtering. Post-hoc spectral processing was performed using
Spectrum (Perkin-Elmer Corporation, Norwalk, CT) when necessary.
Data analysis and spectral decomposition
Decomposition analysis routines were written using two software
packages. Manual curvefitting was performed using Sigmaplot for
Windows, version 5.01 (Jandel Scientific, San Rafael, CA). Automated
processing of bulk spectra was performed using a routine written for
MatLab, version 4.0, (The Mathworks, Inc., Natick, MA). Using these
software routines, Model A was applied to resolve hc for each of the ten
spectral features and baseline OD (k).
One of the goals of the study was to determine if a non-invasive
measurement of tissue hemoglobin saturation could be computed from
absoφtion spectra obtained from an animal's head. It must be kept in
mind that the term "tissue hemoglobin saturation" is a construct that
refers to the average saturation of hemoglobin of brain tissue expressed
as a percentage of the total hemoglobin. In actuality there is no standard
method for determining the value of this parameter. The term "tissue
hemoglobin saturation" stems from a literature citing that suggests
approximately 75% of the blood volume of tissue is contained within the
venous system. On this basis, it has been assumed by a number of
laboratories that approximately 25% of the hemoglobin within a tissue is
contained within the arterial compartment of the vasculature and that 75%
is contained in the venous compartment. Thus, one may estimate a
theoretical "standard" for comparative puφoses. Realistically, the
proportioning between compartments is probably not constant or uniform.
For the study, a distribution of 10%:90% was arbitrarily selected, rather
than the traditional 25%:75%. This was done because the final
relationships between Hb02 / THb and the standard were nonlinear.
Choosing a 10%: 90% ratio appeared to linearize the relationship.
Herein the term "Standard HbSatbt" refers to a computed value that
is believed to represent the average percentage saturation of hemoglobin
contained within a field that is defined by the migration path of photons
shined into tissue. Standard HbSatbt, computed as 0.1 * HbSatan + 0.9 *
HbSatcven, where HbSatgrt and HbSatcven are the percentage saturation of
hemoglobin in arterial and cerebral venous blood, respectively. Standard
HbSatbt was used to verify the NIR HbSatbt method. The features were
resolved independently using a nonlinear curve-fitting algorithm
constrained to yield only positive values for the peak height. Figure 3 is a
graph showing NIR prediction versus measured tissue hemoglobin
saturation. Figure 3 shows that the ratio of optical attenuation at 920 nm
to the sum of optical attenuations of 920 and 760 nm is linearly correlated
to tissue hemoglobin saturation estimated as a weighted fraction of
arterial and cerebral venous hemoglobin saturation. This figure also
shows the results of fitting 180 absoφtion spectra, collected in 21 dogs,
to constituent Model A. Thirty-seven spectra were collected in
normoxic-normocapnic conditions (NN; PaC02 =31-42 mmHg); 98 during
conditions in which hypoxis-hypoxia was induced (HH; Pao2 8 - 100
mmHg); 26 during hypercapnia (HC; PaC02 > 40 mmHg; 4 during anoxia
(AX; Pa02< 8mmHg; 8 in conditions where vascular hemoglobin
concentration was varied (AH; (Hb)a 9-20 g/dl); and 7 during
hypothermia (LT; rectal temperature <30°C). The correlation observed in
each sorted condition was not different from the correlation established
using pooled data. Figure 3 further shows that the ratio 50 tHb02/THb is
proportional to HbSatbt, over the entire standard HbSatbt range 52.
However, Figure 4 shows when the constituents were split out and plotted
against the standard 48, it was found that both oxyhemoglobin 46 and
deoxyhemoglobin 47 were inversely proportional to tissue hemoglobin
saturation 48. Although this is theoretically possible in special
conditions, (e.g., when the fractional extraction of oxygen across the
brain decreases), the measured extraction fraction indicates such a change
could not occur. It was concluded that the oxyhemoglobin measure,
computed using Model A, yielded an erroneous result. It was reasoned
that such an error could arise from cross-talk between oxyhemoglobin
and another constituent. The most likely candidate for cross-talk is the
secondary deoxyhemoglobin peak that overlaps the oxyhemoglobin
spectrum. In Model A, hc for the secondary peak was resolved as though
the variable was independent of all other variables. In fact, it's value is
related to that of the 760 nm variable, because the two peaks are part of
the same complex spectrum.
The model system was revised to include two rules. First, it was
assumed that the deoxyhemoglobin feature at 907 nm was not
independent of the feature at 760 nm, but rather changed proportionately
with the 760 nm feature. Secondly, the location of the feature was
refined. In a small group of test data, the location was determined to be
937 nm, rather than the previously published 907 nm. Hence, a rule,
h937nm> = ct h760nm, was added to the model to account for the linkage,
where is a proportionality constant taken as the ratio of two specific
absoφtion coefficients at the two wavelengths.
A second change was made in the model system to account for the
shoulder region discussed previously when it was found that the 555 nm
and 577 nm features are better characterized as lorentzian shapes, than as
gaussian shapes. As such, these features probably do not contribute
significantly to the NIR band. However, the features of the ultraviolet
band appear to be of such magnitude that they may still be recognizable
in and contribute tilt to, NIR band spectra. Thus, it was decided to
emulate the ultraviolet features using a singular gaussian-shape having a
center location of 391 nm and a FWHM of 150 nm. Although 391 nm is
well above the known location of the ultraviolet features, this value
provided the best fit in a test dataset when the center wavelength of a
single gaussian-shaped peak was assumed and it's FWHM resolved as an
independent variable.
The revised model system, Model B, was reapplied to the
previously-collected spectral library. The results are shown in Figures 5,
6, and 7. Figure 5 shows recovered optical attenuations due to different
hemoglobin constituents - pooled results from fitting to Model B. To
prevent cross-talk between overlapping oxy- and deoxyhemoglobin peaks
in the 920-940 nm region, the deoxyhemoglobin peak at 937 nm was
linked to the deoxyhemoglobin peak at 760 nm. In this fitting result, the
linkage forces a solution where attenuation of 937 nm is 0.6 times
attenuation at 760 nm. The effect on the recovered oxyhemoglobin
attenuation 60 is such that a reduction in this parameter value is observed
as tissue hemoglobin saturation is reduced. In contrast, deoxyhemoglobin
62 rises.
Figure 6 shows the ratio 64 formed as attenuation due to
oxyhemoglobin to the summed attenuations due to oxy- and
deoxyhemoglobin is a linear function 66 of standard tissue hemoglobin
68. The effect of linking the primary and secondary deoxyhemoglobin
peaks and forcing a common solution, is to steepen the slope of the
correlation.
Figure 7 shows the Model B prediction of HbSatbt- Tissue
hemoglobin saturation 70 can be predicted from recovered optical
densities at 920 and 760 nm (see Eq 5). A prediction equation of the
form
OD920nm - C1(OD920nm + OD 60nm)
can be written. The coefficients C] and c2 are the intercept and slope,
respectively, of the tHb02/THb ratio to the standard tissue hemoglobin
saturation plot. The values of c and c2' used to adjust the correlationship
to the line of identity, were 0.4284 and 0.0046, respectively, for this
dataset. In a population of 21 animals, the standard error of prediction if
6.5%. Prediction error is least at tissue saturations greater than 30%.
Below this level the error of prediction is increased, but it is presently
unclear as to whether the fault lies with the model, or with the estimate
used as the standard of comparison.
Referring again to Figure 5, Model B yielded relationships for
deoxyhemoglobin and oxyhemoglobin that were directionally correct,
i.e., when tissue hemoglobin saturation 58 is high, tissue oxyhemoglobin
60 is high and tissue deoxyhemoglobin 62 is low, and when tissue
hemoglobin saturation 58 is low, tissue oxyhemoglobin 60 is low and
tissue deoxyhemoglobin 62 is high. The sum of the two hemoglobin
parameters was inversely related to tissue hemoglobin saturation during
hypoxic conditions and directly proportional to tissue hemoglobin
saturation during hypercarbia (data not shown). This is consistent with
hypoxia induced vasodilation and hypercarbia induced vasodilation;
given that both are initiated from the normoxic-normocarbic state. As
was observed using Model A, Model B yielded an overall excellent linear
relationship between the ratio of the 920 nm constituent to the sum of the
920 nm and the 760 nm constituents (tHb02/THb).
Figures 8A and B illustrate the best and worst fit cases. The left
panel, Figure 8 A, illustrates a correlation between Model B predicted
tissue hemoglobin saturation 74 and the standard of comparison 72 for an
animal with a large index of correlation. The right panel, Figure 8B,
illustrates the worst fit encountered. Overall, the standard error of
prediction of 6.5 was realized.
In a subset of these experiments, optode separation distance was
varied and the relative magnitude of absorbency in the deoxyhemoglobin,
oxyhemoglobin, and water features expressed against standard tissue
hemoglobin saturation (Figure 9). As expected from the modified Beer-
Lambert relationship, and as shown in Figure 9 the optical attenuation 76
attributed to each recovered constituent 78 is a linear function of optode
separation distance 80. This strongly supports the contention that hcis a
composite parameter, dependent in part upon pathlength (see Eq 2). As
such, the absolute concentration of the recovered constituent can be
calculated using a differential pathlength factor and appropriate molar
extinction coefficient. In Figure 9, hc for water (triangles),
oxyhemoglobin (diamonds), and deoxyhemoglobin (squares) is plotted
against the distance between the optodes at the time of measurement.
Regression analysis shows that the y-axis intercept each correlationship is
not different from the x,y axis intercept.
An advantage of using a ratioing method to compute tissue
hemoglobin saturation is illustrated in Figure 10. Tissue hemoglobin
saturation measurements are relatively insensitive to the distance between
optodes. Because tissue hemoglobin saturation is computed as a ratio,
pathlength effects contained by in the numerator and denominator of the
relationship tend to cancel. The method provides a measure that is
optrode separation distance independent, i.e., accurate measurement of
optrode separation does not factor into the overall measurement error. In
the population of dogs studied, tissue hemoglobin saturation measured
using NIR technique during normoxic, normocarbic, normotensive
conditions was 60.0% YSD 11.9% and did not significantly differ from
standard tissue hemoglobin saturation (58.9% + SD 9.1%).
An alternative and preferred model, Method C, for facilitating the
spectral decomposition used to compute brain tissue hemoglobin
saturation includes several revisions to the model used for Model B.
First, the assignments of the 920 nm and 937 nm absorbencies are
reversed. In Model B, the absorbency at 920 nm is attributed to
oxyhemoglobin and the absorbency at 937 nm is attributed to a secondary
deoxyhemoglobin peak. In addition, the FWHM for the two absorbencies
is assigned values of 125 nm and 41 nm, respectively. However, after
considering results obtained from fitting whole blood spectra, in the
alternative and preferred model, these two absorbencies are reassigned to
deoxyhemoglobin and oxyhemoglobin, respectively. In addition, a 41 nm
FWHM is now applied when analyzing spectra obtained from either cat
or dog. The FWHM used to resolve the 937 nm absorbency depends on
the animal species from which the spectra were collected. For cat
spectra, a value of 175 nm yields the best fit to experimental data. For
dogs, the fit is optimum when 130 nm is used to describe the peak. The
physical basis for different FWHM requirements in the two species is
unknown at this writing. These changes necessitate that the linkage
between the 760 nm absorbency and the 937 nm absorbency also be
revised, because the latter absorbency now represents oxyhemoglobin,
rather than deoxyhemoglobin.
Second, the proportionality factor coupling the primary and secondary
deoxyhemoglobin absorbencies is changed. Since the 760 and 920 nm
absorbencies now both represent deoxyhemoglobin, they are coupled in
the revised model using a proportionality factor valued at 1.09 for cats
and 1.175 for dogs. The values were obtained by re-fitting the original
data.
Third, the absorbency at 875 nm is now attributed to lipid.
Originally, the absorbency requirement at 875 nm was considered to be a
secondary peak in cytochrome c oxidase. Measurements taken from
bacon 86 and from the white matter 88 of cat brain now identify a lipid
contribution at this location. The absoφtion spectra shown in Figure 11
were collected by transilluminating a 1 cm sample chamber partially
filled with either fat from bacon, or white matter collected from cat brain.
The spectra have been corrected for water content by differential
spectroscopy. Absoφtion features are notable at 879, 928, and 1038 nm
in both spectra. The FWHM are listed as determined by curve-fitting
each spectrum. The 928nm feature 90 in white matter is small in
comparison to the feature 92 observed for bacon. It should be noted that
the 879 nm features of both fat and cerebral white matter have smaller
FWHM than are needed to resolve tissue spectra. Thus, other
constituents probably overlap in this region. Possible candidates for this
region are: a secondary peak of cytochrome oxidase, or a peak of
cytochrome B.
Fourth, with the current wavelength and FWHM assignments, it is
found that the coefficients, Cj and c2, are not required. Tissue hemoglobin
saturation is computed as a simple ratioing of the two absorbencies:
NIRHbSatbt = 100 x OD937nm / (OD760nm + OD937nm). (Eq 4)
After revising the 920 and 937 nm peaks, the coupling factor
between the two deoxyhemoglobin absorbencies located at 760 nm and at
920 nm also must also be re-determined. In the re-fitting process it was
determined that the coupling factor interacts with the FWHM used to
resolve the 937 nm oxyhemoglobin peak. This interaction is a result of
the overlapping nature of these two absorbers. Increasing the value of the
coupling coefficient assigns a larger portion of the oxyhemoglobin
component to the secondary deoxyhemoglobin component, thus causing
the slope of the relationship between oxyHb(937nm THb and measured
HbSatbt to become greater. This can be offset in part by increasing the
breadth of the oxyhemoglobin peak. This inteφlay allows a solution for
tissue hemoglobin saturation that is not dependent upon correcting
coefficients, such as Ci and c2. Figure 12 shows the result of fitting the
fraction OD937nm / (OD760nm + OD937nrn) 94 to standard HbSatbt yields a fit
that lies on the line of identity and has zero bias. Also, it is seen that
Model C correlates the ratio of oxyhemoglobin 94 to total hemoglobin 96
over a range of zero to 1 , thus eliminating the need for the adjusting
coefficients C\ and c2. This relationship has a zero bias, but tends to yield
a somewhat larger prediction error. Unlike Model B, error appears to
correlate inversely with hemoglobin saturation level and at very low
saturation, a solution for oxyhemoglobin cannot be resolved.
The consequences of changing the coupling factor on the other
variables of the model are not obvious. The constituent most affected by
the change is total hemoglobin. In Model B, the best-fit solution to the
hemoglobin components 98 showed an increased absorbency in the
combined hemoglobin forms 102 during hypoxic-hypoxia and during
hypercapnia (Figure 13 A). This observation is consistent with the
consensus of other reports. The revised model does not yield a similar
result (Figure 13B). It suggests that attenuation attributable to
hemoglobin 102 either remains unchanged, or decreases. (Figure 13B).
The fitting of spectra 98 and 100 for models B and C, respectively,
obtained during hypercarbia remain consistent between the two models.
In both cases hemoglobin attenuation 102, and by inference, cerebral
blood volume, increases. The different results reported by the two
models may be based upon Model B's inability to fully quantify
oxyhemoglobin. Because oxyhemoglobin was only partially resolved, the
rise in deoxyhemoglobin during hypoxia in combination with the
disproportionate representation of the oxyhemoglobin fall, yielded an
apparent rise in total hemoglobin. The physiological literature, currently,
does not offer an explanation for these differences. In support of the
findings from the revised Model C, there is now a suggestion that the
increase in blood flow during hypoxia is mediated through changing
blood flow velocity, rather than volume changes. The data are consistent
with this hypothesis.
Referring again to Figures 13 A and B, the linkage needed to
stabilize cross-talk between oxy-and deoxyhemoglobin biases the total
hemoglobin parameter as an index of blood volume. Hypercapnia
increases tissue hemoglobin saturation, because, by increasing cerebral
blood flow, oxygen delivery is increased with respect to oxygen
consumption. In hypoxia, the resulting decrease in oxygen delivery to
tissue stimulates a compensatory increase in cerebral blood flow that
serves to buffer the change. In both instances, the rise in cerebral blood
flow would be expected to be accompanied by corresponding increases in
cerebral blood volume, and total hemoglobin content of the tissue. In
both Figures 13A and 13B above, hypercapnia was the predominant
cause for tissue hemoglobin saturations lying above 58%, and hypoxia
was the predominant cause for tissue hemoglobin saturations lying below
58%. Thus, one expects a concave relationship with a minima at the
normocapnic-normoxic value, 58%. This was observed when Model B
parameters were used to recover constituent attenuations. However, by
removing the artifact of cross-talk between oxyhemoglobin and
deoxyhemoglobin via the 760 to 920 nm linkage, it is now seen that a
tilting is introduced into the relationship such that hypoxia is not
accompanied by blood volume increases.
The method of the present invention has application to other types
of tissue besides brain tissue. Figures 14 and 15 illustrate the general
nature of the model. Figure 14 shows an absoφtion spectrum 104
collected from the head of a cat and a Model C fit 106 to such spectrum.
The overlay of the Model C fitted absoφtion spectrum 106, was
reconstructed from the absoφtion coefficients that were recovered during
the fitting process. Brain spectra contain two notable features: a peak
centered around 760 nm that is attributable to deoxyhemoglobin, and a
peak centered around 972 nm that is attributable to water. Decomposition
analysis, recovers information on eight additional constituent features,
and in the process yields an excellent fit to the overall absoφtion
spectrum. While the model was developed to decompose brain spectra, it
works equally well when applied to absoφtion spectra collected from
other tissues.
Figure 15 shows an absoφtion spectrum 108 collected from human
gastrocnemius muscle and a Model C fit 110 to such spectrum. Despite
the shape differences between these two spectra (brain and skeletal
muscle), Model C was able to decompose spectra from skeletal muscle
because the constituents of the two tissues are similar. The spectral
features used to compute tissue hemoglobin saturation again include
oxyhemoglobin, deoxyhemoglobin, water, lipid, cytochrome oxidase and
a component for characterizing light loss due to scattering. Thus, it is
expected that decomposition analysis will provide much additional
information about hemoglobin saturation, hemoglobin content, lipid
content, cytochrome oxidase redox state, and water content in a variety of
organ specific tissues. In addition to the features seen in brain tissue
absoφtion spectra, absoφtion spectra contains a prominent triglyceride
peak centered around 928 nm. Because the attributes of lipids and fats
are included, Model C is also able to recover constituent information
from this tissue. Hence, the method appears to be generally applicable
across a variety of tissues and, it is expected, will find application in
studies yet to be conceived.
Although the present invention has been described in terms of a
particular embodiment, it is not intended that the invention be limited to
that embodiment. Modifications of the disclosed embodiment within the
spirit of the invention will be apparent to those skilled in the art. The
scope of the present invention is defined by the claims that follow.