Novel, Noninvasive Multispectral Snapshot Imaging
System to Measure and Map the Distribution
of Human Retinal Vessel and Tissue
Hemoglobin Oxygen Saturation
Kim Alexis Firn,
Bahram Khoobehi
Kim
Alexis Firn, Tulane
University, 6823 St. Charles Ave., New Orleans, LA 70118, USA
Bahram Khoobehi, Louisiana State University Health Sciences Center, 2020
Gravier St., New Orleans, LA 70112, USA
Correspondence to: Bahram Khoobehi,
PhD, Louisiana State University Health
Sciences Center, 2020 Gravier St., New Orleans, LA 70112, USA
Email:
bkhoob@lsuhsc.edu
Telephone: +1-504-568-2385
Fax: +1-504-568-2666
Received: June 5, 2015
Revised: July 26, 2015
Accepted: August 1, 2015
Published online: September 3, 2015
ABSTRACT
AIM: To design and implement a snapshot imaging system capable of mapping
oxygen saturation of retinal vessels and tissue clinically for the first time.
MATERIALS AND METHODS: Our image-splitting design is attached to the imaging
portions of a fundus camera. A relay subsystem and reimaging system convert
reflected white light from the eye into seven monochromatic images
simultaneously. Our algorithm uses intensity information at each of these discrete
wavelengths to approximate the area between the oxy- and deoxyhemoglobin
spectral curves, which has been shown to be proportional to oxygen saturation.
We used MATLAB to convert this information into color-coded oxygen saturation
maps of the optic nerve head. We validated our system by using it on a model
eye with a capillary of known blood oxygen saturation, then used it In vivo
to obtain quantitative values of human oxygen saturation for veins and tissue.
RESULTS: By collecting seven images simultaneously with one
snapshot, our system is the first to document and reproducibly map oxygen
saturation of the retinal vessels and tissue. Oxygen saturation values are
color-coded, with oxygen-rich arteries in red, oxygen-poor veins in blue, and
intermediate tissue yellow-green. Our system’s calculations agree with
co-oximeter readings of blood in a model eye capillary mimicking a vessel. The
quantitative calculations from the healthy subject, as well as the qualitative
oxygen saturation maps, show oxygen saturation levels consistent with those in
the literature.
CONCLUSION: This system is clinically valuable for diagnosing and
monitoring diseases affecting ocular oxygen saturation, such as glaucoma,
diabetic retinopathy, age-related macular degeneration, and others. Our
reliable method is capable of documenting tissue oxygenation throughout disease
progression.
© 2015 ACT. All rights reserved.
Key words: Multispectral;
Oxygen Saturation; Snapshot; Noninvasive; Hemoglobin; Retina
Firn KA, Khoobehi B. Novel, Noninvasive Multispectral
Snapshot Imaging System to Measure and Map the Distribution of Human Retinal
Vessel and Tissue Hemoglobin Oxygen Saturation. International
Journal of Ophthalmic Research 2015; 1(2): 48-58 Available from: URL:
http://www.ghrnet.org/index.php/ijor/article/view/1238
INTRODUCTION
Pathologic conditions in the retina and optic nerve head (ONH) can
cause vision loss and blindness. Both structures have a high demand for oxygen,
and loss of the normal oxygen supply through vascular insufficiency is believed
to play an important role in diseases affecting the retina and ONH. Hypoxia of
the retina and ONH is believed to be a factor in the development of ocular
vascular disorders, such as diabetic retinopathy (DR), arteriovenous occlusion,
and glaucoma. The ability to obtain relative measurements of oxygen saturation
in the human ocular fundus could aid in the diagnosis and monitoring of these
and other disorders. For example, measurement of changes in retinal and ONH
oxygen saturation under controlled conditions could establish relationships
among oxygen consumption, blood sugar levels, and vascular autoregulatory
function in DR. Moreover, the assessment of oxygenation in the ONH may
facilitate early detection of the onset of glaucoma, a disease in which timely
diagnosis is crucial for effective treatment.
Collection of
information about oxygenation levels in the visual system has been previously
carried out with various methods, including physically invasive measurements of
oxygen tension (PO2) in the optic nerve head (ONH) using O2-sensitive
microelectrodes inserted into the eye[1-5]. This method allows
relatively accurate determination of PO2 determination in three dimensions, but
its invasive nature limits it to animal models and precludes human clinical
application. Injection of a phosphorescent dye has also been used to study PO2
in the retinal and choroidal vessels, as well as the microvasculature of the
ONH rim[5-7]. However, this technique has not been approved for use
in humans yet. A non-invasive spectral imaging method shows promise for human
clinical use. This can be a powerful tool for identifying retinal hypoxia
associated with established stages of ocular disease, such as DR. Current research
also seeks to document the oxygenation trend with increasingly severe glaucoma.
Spectral imaging studies of retinal disorders indicate interest in developing
oximetry methodology.
Khoobehi[8]
and Beach[9] have previously adapted non-invasive hyperspectral
imaging, which is an advanced type of spectral imaging, for recording oxygen
saturation data from the primate retina. This technique, in conjunction with
specialized algorithms to compute blood oxygen saturation values, was able to
map relative values of saturation over the retina and optic disc. The technique
was then employed in a study to determine saturation responses to elevated
intraocular pressure (IOP) and oxygen breathing in a primate model for
glaucoma. Results obtained by this method showed that even subtle changes in
the saturation response were detected as IOP was raised. Images were acquired
with a fundus camera (TRC-50vt; Topcon, Tokyo, Japan) that has a lens and a
c-mount through the vertical path of the camera. An imaging spectrograph and a
digital camera were used to obtain hyperspectral images. Khoobehi[8]
used images with 512 spatial points and 256 spectral bands for sufficient light
sensitivity and spectral resolution to monitor oxygen-dependent spectral
changes in vessels. Each frame of the resulting data cube contains the spatial
(x) and spectral (λ) axes for each line of the image. Successive lines form the
z-axis in the stack of frames. Rotating the stack of images, by interchanging
the z and λ axes, gives the “band-sequential” hyperspectral image cube. Each
frame then contains a two-dimensional spatial image at a discrete wavelength.
Scanning the
intensity profile along the z-axis of selected image pixels produced spectral
curves for areas corresponding to arteries, veins, and tissue. These spectral
curves are used to map relative oxygen saturation and produce relative
saturation indices, but the scanning nature of the system requires several
seconds to collect the necessary data. The typical duration of involuntary
saccades is about 100 milliseconds or shorter and depends on the frequency of
eye movement. Further, during the several seconds it takes to scan the eye,
lighting conditions may change and unpredictably complicate the process of
acquiring image data. Therefore, as a result of the sequential nature of the
hyperspectral image collection, the eye must be immobilized for the duration of
the imaging scan. This impedes the imaging procedure and often requires
implementation of special means to prevent the cornea from drying. Although we
successfully recorded oxygen saturation from immobilized retinae in
anesthetized animal studies, our hyperspectral method is not rapid enough for
use in a human clinical setting.
Conventionally-implemented multispectral imaging requires reconstruction
of the final image from the individually obtained spectral images into a
spectral data cube, or composite image, every portion of which contains the
spectral information about the object. This technique involves tunable spectral
filtering[11] or Fourier transform spectrometry[12].
Spectral and spatial multiplexing has low optical throughput[13].
Further, it is only applicable to temporally static phenomena.
In vivo
biomedical imaging of the retina requires snapshot or time-resolved operation
and high optical sensitivity[10]. Bulk-optic[14] and
fiber-optic[15] high spectral resolution snapshot imaging systems
require rearranging a two-dimensional image into a one-dimensional array before
being converted to a spectral data cube by the computed tomographic imaging
spectrometer. The one-dimensional image input is spectrally dispersed at the
detector by the diffractive optical element. The data cube is formed when
computer algorithms invert this data[16,17]. Image-plane
spatio-spectral coding is required to recover the image[18].
Inversion is often accompanied by noise amplification, a drawback of this type
of system.
A system is
needed to overcome the temporal barrier to human clinical tissue oxygenation
analysis posed by scanning systems without the use of injections and to avoid
the inherent complications of bulk-optic and fiber-optic systems. Therefore,
the dual-wavelength snapshot imaging system was developed and used by the
groups of Beach, Hammer, Hardarson, and Tiedman[19-24]. They used
red and green wavelengths for oxygen-sensitive and insensitive monochromatic
images. They succeeded in splitting the reflected light to record pairs of
retinal images side by side on a digital camera or on two different cameras
simultaneously. The green wavelength was set where the oxy- and deoxyhemoglobin
curves crossed with equal and opposite slopes, which kept the reference image
insensitive to oxygen saturation. The red wavelength was set near the maximum
of the oxy-deoxyhemoglobin difference spectrum for high sensitivity. This
technique is capable of human clinical use. Our group has used this technique
successfully to document oxygen saturation of arteries and veins in DR[25].
The high spatial resolution of this system, which can image the entire eye,
comes at the expense of low spectral resolution. Due to having information at
only two wavelengths, this system is not capable of documenting oxygen
saturation of the tissue. The area between the curves of oxy- and
deoxyhemoglobin obtained in hyperspectral scanning systems is proportional to
oxygen saturation of tissue.
Schweitzer and Hammer’s groups at
the University of Jena (Jena, Thuringia, Germany) have developed systems that
obtain one line of a hyperspectral image, and reported retinal oxygen
saturation[26,27]. This Jena imaging spectrograph obtained optical
properties from the tissue environment of the retinal vessel and fundus
background[28,29]. A method of tissue analysis that can map the
entire ONH with a single snapshot is needed since this method can only measure
tissue oxygenation for a single line, and the hyperspectral systems can measure
the tissue oxygenation for the entire ONH only in anesthetized subjects.
Once these
drawbacks are addressed, spectral imaging could be a powerful tool for
identifying retinal hypoxia that is associated with established stages of DR,
glaucoma, central or branch vein occlusion, and age-related macular
degeneration if it were quick enough to be applied clinically. For this reason,
we have now developed a prototype multispectral imaging system fast enough for
clinical application, using a more straightforward approach. The algorithm used
to measure oxygen saturation from these seven images is deduced from the whole
reflected spectrum, which we learned from animal studies using the slow system.
Our non-invasive snapshot imaging system is based on seven-wavelength oximetry.
This optical instrument is capable of mapping oxygen saturation of the retinal
vessels and tissue in the entire ONH by recording seven monochromatic images
simultaneously onto a large chip with high dynamic sensitive digital cameras,
integrated with a single optical connection to a fundus camera. Our oxygen
saturation algorithm is applied to the information from these seven images to
calculate oxygen saturation of retinal vessels and tissue and create entire ONH
maps clinically for the first time.
This paper
reviews our efforts to develop this snapshot imaging system. We will describe
our multispectral imaging system and oxygen saturation algorithm. This imaging
system is capable of separating the white light reflected from the retina into
seven discrete wavelengths. The seven monochromatic images created from
separating this scattered light are analyzed for light intensity, an indicator
of the hemoglobin light absorption in the ocular structure being analyzed. The
light absorbance of a certain structure in the eye, such as a vessel or tissue
area, at each of these seven wavelengths is directly proportional to the
relative hemoglobin oxygen saturation. Our algorithm has been coded into a
MATLAB program capable of transforming images into color-coded oxygen
saturation maps of the eye. We validate this imaging system by applying it to a
model eye with blood of known oxygen saturation. Then, we report our
preliminary quantitative human results for a healthy subject.
METHODS
Multispectral
Imaging Instrumentation Design
We designed and created a snapshot multispectral imaging system. The
system can obtain seven monochromatic images simultaneously. The prototype
capable of splitting images is attached to the imaging portions of a
commercially available fundus camera (Zeiss FF450 IR). Traditionally, a
patient’s eye image is captured through a fundus camera at the pupil (A).
The schematic
of the attachment to the fundus camera is shown in Figure 1, which includes a
relay subsystem and has an entrance pupil (B). The entrance pupil (B) is the
same as the exit pupil (A) of the fundus camera. A reimaging subsystem (RI)
includes a bandpass filter array (D) and an array of secondary objectives (E),
which is positioned at an exit pupil (F) of the relay subsystem. Each of the
individual objectives of the array (E) focuses light on a 2-D focal plane
detector to produce individual spectral images of the retina of the eye. In
this particular device, we made a prototype of a reimaging subsystem (RI) with
seven lenses and bandpass filters.
In reference
to Figure 1, the five elements (lenses) of the relay subsystem can be divided
into three groups. The first (positive) group with the primary objective 1, the
second (correction) group with lenses 2, 3, and 4, and a third (positive) group
with lens 5. The first and third groups operate to reimage and expand the exit
pupil (B) of the fundus imaging device into the exit pupil (F) of the relay
subsystem (RI). The second group is configured to correct optical aberrations
that would otherwise render the optical quality of the images unusable. An
image frame containing individual spectral images is further developed with a
computer processor and appropriate data processing algorithms to produce a
resulting aggregate multispectral image frame.
Figure 2 shows
detail of the RI and the images formed. The transverse dimensions of the exit
pupil (F) are preferably slightly larger than those of the RI to avoid
vignetting and/or shadowing effects. There are seven different bandpass filters
that have been used (3 oxygen-sensitive at 542, 560, and 577 nm; 4 isosbestic
at 522, 548, 569, and 586 nm). As shown in Figure 3, each of the individual
secondary objective lenses of the RI is individually repositionable via an
associated elongated lens barrel. This provides a means for readjusting the
positioning of the given secondary objective lenses along the optical axis to
obtain a highly focused corresponding individual image frame. The lens barrels
are assembled in a single mount that utilizes a barrel-locking mechanism to
maintain the position of the lens barrels after the focus adjustment. We used
this snapshot system to image the fundus. The outcome is seven identical
monochromatic images recorded in a 32 MB chip with 16 bit high dynamic
resolution.
We designed a
more compact layout without changing the optical path of the overall system,
shown in Figure 4. This configuration makes the system compact and facilitates
its integration with the fundus camera. In Figure 4, the lenses are labeled in
the same fashion as in Figure 1. Several adjustable folding mirrors are
utilized (labeled M in Figure 4) and lens 3 of the relay subsystem is removed
to accommodate the preferred optical layout. Array D contains seven optical
filters with corresponding bandwidths of about 4 nm centered at 522 nm, 542 nm,
548 nm, 560 nm, 569 nm, 577 nm, and 586 nm, respectively. The image frame
contains seven individual images retaining individual spectral responses of
corresponding individual bandpass filters. This frame is further split into
individual images that are processed and recombined, with the use of a computer
processor, into a final composite multispectral image.
Optical design
parameters for this snapshot multispectral imaging system are summarized in
figures 4 through 8. Figure 4 illustrates the geometrical layout of the imaging
system, while Figure 8 presents transmission characteristics of the optical
filters of the array D. Figures 6 and 7 illustrate spot diagrams associated
with an imaging system having a full aperture and that of the imaging system
with a segmented aperture, respectively. As expected, the diffraction limit of
the resulting image of the segmented aperture system is slightly lower than
that of a full-aperture system where the exit pupil of the relay sub-system is
not spatially segmented. These diffraction limits are 12.21 microns and 3.66
microns, respectively. Calculations are obtained from an optical software
design program (Zemax) and are shown in Figure 5. Reduction in spatial
resolution does not affect this system’s use in ophthalmologic imaging, where
the spatial resolution of the eye remains a limiting factor.
A typical set
of seven monochromatic images of a human fundus obtained from the system are
shown in Figure 9. These images have a 16 bit original dynamic range. However,
the screen limits the viewing capacity to 8 bit resolution. The resulting
images are arranged from 522 nm at 11:00 to 577 nm at 1:00 in a
counterclockwise fashion with the 586 nm image in the center.
Algorithm
A hyperspectral algorithm was developed in our
laboratory for the quantification of oxygen saturation in retinal artery and
ONH tissue[8,9]. Figure 10 shows the whole spectrum for oxy- and
deoxyhemoglobin obtained from reflected artery and vein overlaying the ONH[8,9].
This whole spectrum was obtained from the ONH of a nonhuman primate. The
reflectance oxyhemoglobin spectra in this figure was achieved by letting the
animal breathe pure oxygen. The reflectance deoxyhemoglobin spectra was
obtained from desaturating oxyhemoglobin by applying a suction cup to the eye
and raising IOP to 45 mmHg. At this high IOP, perfusion stops at the fundus for
a short period of time and blood and tissue become desaturated[8,9].
Animals had to be anesthetized to immobilize the eye, which was held open for
the duration of an imaging session. Any involuntary eye movement that might
occur in the conscious animal would compromise the data.
As seen from
the figure, there are four points where the oxy- and deoxyhemoglobin spectra
intersect. These four points are at oxygen-insensitive (isosbestic)
wavelengths. In other words, these four points have the same absorption and
reflection for oxy- and deoxyhemoglobin. Previously, we developed an algorithm
that measures oxygen saturation from the whole spectrum. Briefly, we proved
oxygen saturation is proportional to the areas trapped by the spectrum and the
lines that connect the isosbestic points to each other. As seen in Figure 6,
the whole spectrum has three minimum or maximum points, which are
oxygen-sensitive. Using these three oxygen-sensitive points (542, 560, and 577
nm) and four isosbestic points (522, 548, 569, and 586 nm), we can construct
three triangles which have areas proportional to oxygen saturation, as shown in
figure 11. We constructed an algorithm based on these seven discrete
wavelengths from the whole hemoglobin spectrum. Discrete spectral curves from individual
pixels within the border of each structure can be constructed (Figure 11).
These curves are then fit to a linear combination of reference curves obtained
from fully saturated and desaturated red cell suspensions. Before performing
the curve fit, the recorded spectrum is transformed by the method of Hammer et
al. to remove influences of nonhemoglobin light absorption and light scattering[30].
This transformation corrects the recorded curves at three isosbestic
wavelengths (522, 569, and 586 nm) to match the corresponding intersections of
the reference curves for oxygenated and deoxygenated blood. Saturation is
measured at 560 nm, which is a maximum in the difference spectrum. We use the
same procedure to test the calibration of red cell suspensions. Then we
determine saturation by least square fits to oxygenated and deoxygenated
reference curves from red cell suspensions.
We find the
oxygen saturation obtained as described above is proportional to the summation
of three triangles normalized to the intensity of the incident light. These
triangles are constructed with four isosbestic and three oxygen-sensitive
wavelengths.
Curve fits are
performed with a Windows software package (MathGrapher 2.0; Springfield Holding
b.v., Noordwijk, The Netherlands). Reference spectra of saturated (Ssat) and
desaturated (Sdesat) red cell suspensions are fit to transformed retinal blood
spectra (S) using fitting parameters A and B with an additive term (C), as in
Eq. (1):
S = A × Ssat +
B × Sdesat + C. (1)
Percent oxygen
saturation is determined by expressing fitting parameters as in Eq. (2):
%Sat = 100 ×
A/(A + B), (2)
Where A and B
correspond with best-fit coefficients for oxyhemoglobin and deoxyhemoglobin
contributions as defined by Eq. (1).
To calibrate
the system In vivo, we are going to use the approach used by Hammer et
al[30] to correct for light level and tissue pigmentation.
Hammer’s
method[30] uses a series of linear transformations to mitigate
non-hemoglobin absorption and tissue scattering. In order for this procedure to
work, reference spectra of 100 percent oxygenated blood and fully desaturated
blood need to be known. The spectra are forced to have equal end points at the
isosbestic wavelengths of 522 nm and 586 nm. This is done by first making a
linear function connecting the start and end points of isosbestic wavelengths
(522 nm and 586 nm) for the references and vessel spectrums, g(λ) and f(λ)
respectively. The measured spectral curves (Mλ) are then transformed by an
equation for all wavelengths to correct for light level. The next step is to
force the spectra to go through the same point at 548, which is done using an
equation to correct for tissue pigmentation. The final calculation involves the
relative intensity of the transformed spectrum at 560 nm (oxygen-sensitive
wavelength) to the reference spectrum[30]. Hammer used this
technique for the whole hemoglobin spectrum. We followed the same procedure to
obtain seven wavelength multi-spectral images.
In vitro study with a model eye
To validate our results, we prepared differentially oxygenated blood
samples and placed them in a model eye to be imaged by our system. The model
eye was designed to mimic the optics of the human eye, consisting of a
60-diopter lens and a chamber filled with circulating water. A 50-μm tube
connected to a circulating pump was placed at the image plane of the lens,
acting as a model retina. The differentially saturated solutions were then
individually run through the tube and imaged by our multispectral imaging
system.
After the
plasma was removed from the blood sample, it was re-suspended in an aqueous
buffer solution. We bubbled oxygen gas through the solution bringing it to
complete 100 percent saturation. We then bubbled nitrogen gas through the
solution for different amounts of time to get set levels of oxygen saturation.
Using a co-oximeter (Instrumentation Laboratory Co-Oximeter Model 682), we
measured oxygen saturation of each sample.
In vivo sample of a healthy human subject with quantitative
values
This study was approved by the Institutional Review Board of LSU. It
adheres to the tenets of the Declaration of Helsinki. The subjects provided
informed consent before participation in the study and went through standard
study protocol. After consenting to be in the study, the subjects answered a
questionnaire about medical history.
The healthy
subject imaged and analyzed in this study is a 54-year-old Caucasian male with
olive skin pigmentation. Figure 12 shows the area where measurements were taken
from the healthy subject for oxygen saturation values. As indicated, we divided
the image into four concentric circles. Area 1 is located on the perimeter of
the inner circle. The numbers of the areas increase radially outwards, with
Area 4 being on the outermost circle. Superotemporal, superonasal,
inferotemporal, and inferonasal areas of both vein and tissue were measured for
each of four distances from the ONH. Those distances were (1) just outside the
ONH; (2) one ONH diameter from the ONH edge; (3) two ONH diameters from the ONH
edge; and (4) three ONH diameters from the ONH edge. For the vein areas, ten
pixels are averaged to get the oxygen saturation reading. For the tissue areas,
80 pixels are averaged. The vein values are calculated by comparing vein
intensity to artery intensity, which we set to 96 percent saturation.
Oxygen
saturation maps
The seven monochromatic images from a single snapshot of our system are
used to create a color-coded oxygen saturation map of the eye. After registration
of the seven monochromatic images, color-coded MatLab figures are generated by
our MATLAB script, as described in previous work[8]. Spatial
resolution is per pixel, but oxygen saturation values are relative. The
reported number is the average of a certain amount of pixels. The oxygen
saturation values are color-coded in a gradient from highest in red to lowest
in blue. The arteries appear red to show high oxygen saturation. The veins
appear blue to show low oxygen saturation. The tissue appears yellow-green to
show intermediate oxygen saturation. We imaged a healthy 35-year-old Caucasian
woman, and we show the seven simultaneously obtained monochromatic images and
the oxygen saturation map for this subject in the results section. Also,
different technicians imaged a healthy 38-year-old Caucasian woman on separate
days to show reproducibility of our oxygen saturation maps.
RESULTS
In vitro study with a model eye
We detailed the creation of a model eye with blood of known oxygen
saturation run through a capillary mimicking a retinal vessel within the model
eye cavity. The results from our imaging system and algorithm’s use on our
model eye are plotted against the known blood oxygen saturation in Figure 13.
There is a linear correlation between oximeter readings and our system’s
calculations.
The x-axis in
the figure indicates the oxygen saturation values of the blood samples that we
obtained from different aliquots. The raw data was converted to oxygen
saturation by our algorithm and plotted on the y-axis. The co-oximeter values
vary linearly with our algorithm's values, with R2=0.984, as shown
in the figure.
In vivo sample of a healthy human subject with quantitative
values
Table 1 indicates average oxygen saturation values for vein regions
corresponding to Areas 1 through 4, respectively. The last row in the table
gives the average saturation from all radii for each region. Table 2 is the
analogous chart for tissue regions. These values are obtained from our
previously mentioned algorithm, with an assumption that the arteries have an
oxygen saturation of 96 percent. Each region’s oxygen saturation is an average
of the values from day 1 and day 7 of oxygen saturation calculation.
To test
reproducibility we repeated the multispectral imaging in two different
sessions, seven days apart, and averaged the results. The values for the vein
changed from 63 ± 4.7 percent to 59 ± 2.9 percent, and the values for the
tissue changed from 71 ± 4.8 percent to 76 ± 4.5 percent. The average value
taken from all vein areas was found to be 61 percent, whereas this average
oxygen saturation value for the tissue was 76 percent. We found that the oxygen
saturations of the superotemporal, inferotemporal, and inferonasal veins were
of similar values, ranging from 62 to 63 percent. That of the superonasal vein,
however, was significantly lower, at 54 percent (p ≤ 0.05). We also found that all of
the oxygen saturations for the tissues in different quadrants of the eye were
quite similar to each other, ranging from 74 to 77 percent.
The percent
oxygen saturations of the superotemporal vein, superonasal vein, and of the
superotemporal tissue all tended to decrease as one moved radially farther from
the ONH, toward the periphery of the image. The inferotemporal and inferonasal
veins, as well as the inferotemporal, inferonasal, and superonasal tissues, did
not show any correlation between distance from the ONH and percent oxygen
saturation.
Oxygen
saturation maps and reliability
The seven images in Figure 9 were converted to a color-coded figure of oxygen saturation, which is shown in Figure 14 for a typical ONH. Images from this system’s clinical use In vivo show that the results are reliable and reproducible. The images, calculations, and color-coded oxygen saturation maps are relatively constant for a given patient on different days as taken by different technicians. The patient whose images are compared here had no known ocular health changes between image acquisition trials. Figure 15 shows two color-coded oxygen saturation maps of the fundus of the same patient imaged on different days by different technicians, exhibiting reliability.
DISCUSSION
Laser doppler flowmetry has shown that various retinal diseases and
treatments are associated with significantly altered blood flow rates. The
establishment of a relationship between altered blood flow and pathogenesis of
diseases and the prospect of diagnosis based on functional changes are exciting
new developments. After background retinopathy is established in diabetic
individuals, adjacent areas of high and low blood perfusion are often
visualized by fluorescein angiography. Areas of low perfusion become hypoxic
and, over time, become sites of neovascularization which leads to proliferative
retinopathy and blindness. Areas of high perfusion subject retinal capillaries
to unfavorable flow and pressure over time. This stress is believed to cause
the microaneurysms and hemorrhages which are first clinical signs of
retinopathy. Structurally compromised retinae may receive a normal supply of
oxygen. However, if inner retinal oxygen consumption is increased during
periods of hyperglycemia, even these areas could experience relative hypoxia[31].
The pathogenic mechanisms leading to adjacent high and low areas of blood flow
are not well understood. Possibly this situation could result from impaired
blood flow autoregulation caused by high blood sugar. Regardless of the origin,
a means of identifying hypoxic retinal tissue would be extremely valuable, both
to identify individuals at risk for severe stages of retinopathy prior to
development of structural pathology and to investigate how oxygen utilization
is implicated in pathogenesis of retinopathy. An ability to detect and correct
abnormal retinal function before any vascular and tissue damage can occur will
greatly improve outcomes and avoid injury. The development of an instrument
capable of oximetry of the ocular fundus enables such detection.
Several
oximetry studies based on the dual-wavelength method of Beach et al[19]
have been reported recently. Harris et al[32] evaluated the
performance of dual-wavelength oximetry by imaging vessels near the optic disc. Crittin et
al[33] reported time resolved measurements of venous saturation
in response to breathing 100 percent oxygen using the dual-wavelength
technique. A commercial version of the dual-wavelength imaging method was
introduced by Oxymap company (Iceland) and it employed the two wavelengths used
by Beach et al[19] earlier. This system splits retinal images
into two separate wavelengths that are recorded on two separate sensors. It
measures the relative optical density from oxygen-sensitive and insensitive
images, and it relates the relative optical density to oxygen saturation. The
system is capable of measuring oxygen saturation in the fundus vasculature, but
not tissue.
Another method
combines visualization of flow and oximetric state using rapid flash sequences[34].
The hypoxic areas shown from this instrument were under areas of retinal
hemorrhage and were not clear in the image; more results are needed to confirm
this approach for clinical assessment.
The above
techniques measure the oxygen saturation of retinal veins and arteries. We
developed a snapshot method capable of taking multiple images simultaneously
using seven wavelengths, determined to be necessary by hyperspectral imaging,
to find the oxygen saturation not only in blood vessels but in the retinal
tissue as well. Our system is the first noninvasive imaging system capable of
documenting the oxygen component of both vessels and tissue clinically.
Hemoglobin oxygen saturation is a biomarker that varies with severity of
several prevalent conditions, such as glaucoma, DR, macular degeneration,
retinopathy of prematurity, retinal artery and vein occlusion, and others. A
quick, easy, affordable diagnostic and disease monitoring technique would be an
extremely valuable clinical tool.
A basic
spectral imaging system includes a detector or a detector array positioned in
the imaging plane of an optical system that images a scene or an object, and a
series of optical filters (e.g., bandpass filters defining different spectral
regions of imaging) intersecting the beam of light delivered to the imaging
plane. Individually obtained spectral images can be taken in a sequence (in a
system having a single imaging channel) or simultaneously (in a multi-channel
system). Imaging with the use of a single-channel system typically requires
changing optical filters one at a time. The final image is then reconstructed
from the individually obtained spectral images into what is sometimes referred
to as a “spectral cube” or a “composite image,” every portion of which contains
the spectral information about the object. In many circumstances, the object
being imaged is not stationary. This complicates imaging with a single-channel
system because sequentially acquired images from one another and require
complicated reconstruction processes. In other cases, such as medical imaging,
it is not feasible to take many images due to patient compliance issues or
limited allotted time. Our snapshot multispectral imaging technique circumvents
these shortcomings by obtaining seven images simultaneously.
The aperture
stop of an optical system is the location where the rays from every point of a
scene cross. It is typically where a fixed or adjustable iris is positioned to
control the amount of light going through the optical system. Its size
determines the f-number of the optical system. The pupil of an optical system
is the image of the aperture stop as seen by the object (entrance pupil) or by
the image (exit pupil). As in the case of the aperture stop, the exit pupil is
where the rays originating from a point of the object cross everywhere in the
pupil. This unique feature enables the pupil to be segmented into smaller
portions, each containing all the spatial information from the object. In order
to construct a multispectral imaging system, a series of objective lenses are
arranged in the exit pupil in a hexagonal pattern with a seventh lens in the
center. Each lens produces an individual image of the scene on the focal plane.
The focal length and diameter of the individual objective lenses determine the
size and spacing of the scene image on the focal plane. Those parameters are
adjusted to optimally fill the focal plane and obtain the highest spatial
resolution. In addition, the objective lenses are preceded by a filter, each
covering a specific wavelength of interest. The final multispectral image is
reconstructed using the individual images contained in the single file.
Our new idea
arises from the realization that the use of a two-dimensional array of
secondary objective lenses positioned to spatially split or segment the
incoming beam at the image plane of the entrance pupil significantly simplifies
the multispectral, multi-channel imaging system. In such a configuration, the
array of secondary objectives performs the beam-splitting role, and there is no
need for a separate beam-splitting component. As a result, folding of the
optical path can be avoided. Additional advantages of this configuration
include simplicity of assembly, modularity, and reconfigurability of the
imaging system.
In
conventional multi-channel systems, precision and symmetry of the positioning
of beam-splitting components in a transverse (with respect to the optical axis
of the system) plane defines the resulting spatial resolution in the image
plane. However, in applications that do not require imaging systems with
maximized spatial resolution or that can employ imaging systems having spatial
resolution below a pre-defined threshold, axially-asymmetric positioning of the
secondary objectives forming multiple images in the imaging plane can be
sufficient. Using secondary objectives of the reimaging subsystem at the exit
pupil to spatially segment the incoming beam allows us to avoid the use of
auxiliary beam-splitting components used in previous imaging techniques[14].
This increases the system tolerance to mechanical misalignments.
Another
advantage of our imaging system is the ability to record a complete spectral
image using a two-dimensional focal plane detector array in a single exposure,
without the need for spatially deviating the image-forming beams from one
another. This greatly helps in the recombination of the individual images into
a single spectral image. This advantage is particularly pronounced in
applications involving imaging of dynamic objects, such as the eye of a
conscious animal.
We built a
system capable of retinal oximetry and we validated it using a model eye. Our In
vitro study shows a linear correlation between co-oximeter readings and our
multispectral method with R2=0.984. The red blood cell suspension in
our In vitro study is free of any tissue pigmentation. Adjusted curves
(seven reflectance intensity values at discrete wavelengths connected to each
other) from red cells with different levels of oxygenation and from
deoxygenated red cells are plotted with curves obtained from retinal arteries
of a normal subject. Then, saturation of all samples were determined at a
wavelength of 560 nm by the method of Hammer[20]. For the oxygenated
red cells with 96 percent oxygenation read by the co-oximeter, our method
agrees within 2 percent of the 96 percent reading.
We draw four
major conclusions from these experiments with our technique. First, the venous
oxygen saturation values are reproducible, since they are comparable to other
investigators' data.However, our superonasal vein results are lower than our
other vein results, which does not agree with other studies. With an increased
study sample, this effect will likely disappear. Since this is a feasibility
study for proof of concept, we only evaluated one eye. A dual-wavelength study
by Geirsdottir et al[35] found the inferotemporal quadrant to
have lower oxygen saturation measurements than other quadrants for both
arterioles and venules. Retinal oxygen saturation is individually variable, and
our subject’s results shows variation from the results of Geirsdottir's group.
On the whole, our data from veins demonstrate that we are recording results
very similar to other human trials[19,36-37]. Hammer and Schweitzer et
al found in their study[36] that the mean venous oxygen
saturation was 65±11.7 percent, which our data corroborates. Stefansson and
coworkers[37] found in their research that the percent oxygen
saturation in retinal veins was 63±5 percent, substantiating our range of
values. Also confirming our results were Beach et al who found that the
percent oxygen saturation of retinal veins in subjects breathing room air
ranged from 54 percent to 62 percent[19]. Confirming our venous
values with other techniques indicates that our spectrum technique leads to
similar results as those obtained by the optical density approach used by
others[19,36-37].
In future
studies, we will calibrate our system to standardize the results with those
from the accepted dual-wavelength method. This will prevent having to assume
arterial saturation is always 96 percent, and allow us to measure differences
in arterial saturation, as well. Our system requires an arterial reference
oxygenation because we did not calibrate our results with an accepted system,
such as the dual-wavelength method. Alternatively, we could have altered
oxygenation externally through oxygen-breathing and monitored percent oxygen
saturation corresponding to our readings, but we have not done this. If the
actual arterial saturation is lower or higher than 96 percent, the vein and
tissue values will appear too low or too high, respectively. We did not apply our
algorithm's values from the model eye study to our in vivo system because the
model eye pipette is not in the same environment as a real vessel. Therefore,
the resulting values are not corrected for real conditions, like pigmentation.
The in vitro model eye study only confirmed our algorithm gives a linear
relationship with spectrometer readings.
Second, our
technique is the first non-invasive imaging method capable of measuring oxygen
saturation of microcirculation, or retinal tissue. The oxygen saturation values
that we measured for the retinal tissue include the signal generated by the
choroid, as well. These two tissues cannot be separated by this technique.
However, we speculate that the signal generated by the choroid is not
significant in this particular subject. The subject had olive skin pigmentation
and the color fundus image shows dark retinal pigment epithelium. Therefore,
the signal generated from the choroid is minimal. This conclusion is supported
by our results, which show that all four areas radiating outward from the ONH
edge are similar. The area just outside the ONH is composed of the nerve fiber
layer, or the fibers converging to form the optic nerve. This area is much like
the actual ONH in that it does not have a background contribution from the
choroid. If all areas show similar oxygenation results, and if area 1 has the
same background as the ONH, then it appears that there is no difference in
background signal affecting our oxygenation results. If areas 2, 3, and 4 had a
significant choroidal contribution, they would all vary from area 1 due to this
effect. The superotemporal tissue region does show a decreasing trend moving
radially outward from the ONH, but if this were due to choroidal contribution,
all regions would show this trend.
Third, our
system is able to map oxygen saturation throughout the entire retina. The
current study is the first to our knowledge to report and map the distribution
of blood oxygen saturation in retinal structures. There was good agreement between
qualitative results obtained by the mapping method algorithm developed by our
group and our quantitative oxygen saturation results. Both the quantitative
results from a healthy subject in Tables 1 and 2 and the qualitative results
from two different healthy subjects in Figures 14 and 15 showed that the order
of high, intermediate, and low saturation in, respectively, the retinal
arteries, retinal tissues, and retinal veins, are as expected. The advantage of
averaging intensity values from pixels in areas of interest, as was done to
generate the oxygen saturation values for veins and tissue areas in Tables 1
and 2, is noise reduction. Significant additive noise appears on multispectral
curves from single pixels. Figures 14 and 15 are generated from such single
pixel recordings and exhibit increased noise.
Fourth, our
results are reliable. As previously mentioned, the quantitative values obtained
from the healthy subject on days 1 and 7 for each region (superotemporal,
superonasal, inferotemporal, or inferonasal) and area (1, 2, 3, or 4) do not
vary significantly. Further, the color-coded oxygen saturation figures mapping
the oxygenation of the eye are reliable, as shown by Figure 15. These methods
for generating oxygen saturation percentages and relative values for
color-coded maps are invaluable clinical tools, and these results confirm that
our method is reliable.
In conclusion,
our reliable method for clinical documentation and mapping of vessel and tissue
oxygen levels is novel and of clinical relevance. Our system overcomes the
inherent drawbacks of previous oximetry methods, and can be used to monitor
disease state noninvasively.
ACKNOWLEDGMENTS
We would like to acknowledge the
support of Louisiana State University Health Sciences Center, Lions Club, and
Research to Prevent Blindness.
CONFLICT
OF INTERESTS
The author has no conflicts of
interest to declare.
REFERENCES
1
Linsenmeier RA, Braun RD, McRipley MA,
Padnick LB, Ahmed J, Hatchell DL, McLeod DS, Lutty GA. Retinal hypoxia in
long-term diabetic cats. Invest Ophthalmol Vis Sci. 1998 Aug;39(9):1647-57.
PubMed PMID: 9699554.
2
Wangsa-Wirawan ND, Linsenmeier RA.
Retinal oxygen: fundamental and clinical aspects. Arch Ophthalmol. 2003
Apr;121(4):547-57. Review. PubMed PMID: 12695252.
3
Lau JC, Linsenmeier RA. Increased
Intraretinal PO2 in Short-Term Diabetic Rats. Diabetes. 2014
Dec;63(12):4338-42. doi: 10.2337/db14-0101. Epub 2014 Jul 15. PubMed PMID:
25028524.
4
Riva CE, Pournaras CJ, Tsacopoulos M.
Regulation of local oxygen tension and blood flow in the inner retina during
hyperoxia. J Appl Physiol (1985). 1986 Aug;61(2):592-8. PubMed PMID: 3745049.
5
Cranstoun SD, Riva CE, Munoz JL,
Pournaras CJ. Continuous measurements of intra-vascular pO2 in the pig optic
nerve head. Klin Monbl Augenheilkd. 1997 May;210(5):313-5. PubMed PMID:
9230497.
6
Riva CE. Noninvasive measurement of
oxygen tension in the optic nerve head. Curr Opin Ophthalmol. 1998
Apr;9(2):56-60. Review. PubMed PMID: 10180515.
7
Shahidi M, Wanek J, Blair NP, Little DM,
Wu T. Retinal tissue oxygen tension imaging in the rat. Invest Ophthalmol Vis
Sci. 2010 Sep;51(9):4766-70. doi: 10.1167/iovs.09-4710. Epub 2010 Apr 7. PubMed
PMID: 20375336.
8
Khoobehi B, Beach JM, Kawano H.
Hyperspectral imaging for measurement of oxygen saturation in the optic nerve
head. Invest Ophthalmol Vis Sci. 2004 May;45(5):1464-72. PubMed PMID: 15111603.
9
Beach J, Ning J, Khoobehi B. Oxygen
saturation in optic nerve head structures by hyperspectral image analysis. Curr
Eye Res. 2007 Feb;32(2):161-70. PubMed PMID: 17364749.
10 Gorman
A, Fletcher-Holmes DW, Harvey AR. Generalization of the Lyot filter and its
application to snapshot spectral imaging. Opt Express. 2010 Mar
15;18(6):5602-8. doi: 10.1364/OE.18.005602. PubMed PMID: 20389576.
11 Hardeberg
JY, Schmitt F, Brettel H. Multispectral color image capture using a liquid
crystal tunable filter. Opt. Eng. 41(10), 2532–2548 (2002).
12 Harvey
AR, Fletcher-Holmes DW. Birefringent Fourier-transform imaging spectrometer.
Opt. Express 12(22), 5368–5374 (2004).
13 Harvey
AR, Beale JE, Greenaway AH, Hanlon TJ, Williams JW. Technology options for
hyperspectral imaging. Proc. SPIE 4132, 13–24 (2000).
14 Weitzel
L, Krabbe A, Kroker H, Thatte N, Tacconi-Garman LE, Cameron M, Genzel R,
Tacconi Garman LE. 3D: The next generation near-infrared imaging spectrometer.
Astron. Astrophys. Suppl. Ser. 119(3), 531–546 (1996).
15 Fletcher-Holmes
DW, Harvey AR. Spectral imaging with a hyperspectral fovea. J. Opt. A, Pure
Appl. Opt. 7(6), S298–S302 (2005).
16 Descour
MR, Volin CE, Dereniak EL, Thome KJ, Schumacher AB, Wilson DW, Maker PD.
Demonstration of a high-speed nonscanning imaging spectrometer. Opt. Lett.
22(16), 1271–1273 (1997).
17 Johnson
WR, Wilson DW, Fink W, Humayun M, Bearman G. Snapshot Hyperspectral Imaging in
Ophthalmology. J. Biomed. Opt. 12(1), 014036 (2007).
18 Wagadarikar
AA, Pitsianis NP, Sun X, Brady DJ. Video rate spectral imaging using a coded
aperture snapshot spectral imager. Opt. Express 17(8), 6368–6388 (2009).
19 Beach
JM, Schwenzer KJ, Srinivas S, Kim D, Tiedeman JS. Oximetry of retinal vessels
by dualwavelength imaging: calibration and influence of pigmentation. J Appl
Physiol. 1999;86:748–758.
20 Hammer
M, Vilser W, Riemer T, Schweitzer D. Retinal vessel oximetry-calibration,
compensation for vessel diameter and fundus pigmentation, and reproducibility.
J Biomed Opt. 2008;13: 54015.
21 Tiedeman
JS, Kirk SE, Srinivas S, Beach JM. Retinal oxygen consumption during
hyperglycemia in patients with diabetes without retinopathy. Ophthalmol.
1998:105:31–36.
22 Hardarson
S, Stefansson E. Oxygen saturation in branch retinal vein occlusion. Acta
Ophthalmol. 2012;90:466–470.
23 Hammer
M, Vilser W, Riemer T, et al. Diabetic patients with retinopathy show increased
retinal venous oxygen saturation. Graefes Arch Clin Exp Ophthalmol.
2009;247:1025–1030.
24 Hardarson
S, Stefansson E. Retinal oxygen saturation is altered in diabetic retinopathy.
Br J Ophthalmol. 2012;96:560–563.
25 Khoobehi
B, Firn K, Thompson H, Reinoso M, Beach J. Retinal arterial and venous oxygen
saturation is altered in diabetic patients. Invest Ophthalmol Vis Sci. 2013 Oct
29;54(10):7103-6. doi: 10.1167/iovs.13-12723. PubMed PMID: 24114546.
26 Schweitzer
D, Hammer M, Kraft J, Thamm E, Konigsdorffer E, Strobel J. In vivo
measurement of the oxygen saturation of retinal vessels in healthy volunteers.
IEEE Trans Biomed Eng. 1999;12:1454–1465.
27 Schweitzer
D, Thamm E, Hammer M, Kraft J. A new method for the measurement of oxygen
saturation at the human ocular fundus. Int Ophthalmol. 2001;23:347–353.
28 Hammer
M, Roggan A, Schweitzer D, Mϋller G. Optical properties of ocular fundus
tissues-an In vitro study using the double-integrating-sphere technique
and inverse Monte Carlo simulation. Phys Med Biol. 1995;40:963–978.
29 Hammer
M, Schweitzer D. Quantitative reflection spectroscopy at the human ocular
fundus. Phys Med Biol. 2002;47:179–191.
30 Hammer
M, Thamm E, Schweitzer D. A simple algorithm for In vivo ocular fundus
oximetry compensating for non-haemoglobin absorption and scattering. Phys Med
Biol 2002;47:N233-N238. PMID: 12361226
31 Tiedeman
JS, Kirk SE, Srinivas S, Beach JM. Retinal oxygen consumption during
hyperglycemia in patients with diabetes without retinopathy. Ophthamology 105:
31-36,1998.
32 Harris
A, Kageman L, Amin C, Migliardi R, Ciulla T, Arend O. A New Spectrographic
Digital Fundus Oxymeter: Effect of Hyperoxia on Retinal Oxygen Tension.
DOG-Jubilaumstagung 2002 Sept.
33 Crittin
M, Schmidt H, Riva CE. Hemoglobin oxygen saturation (So2) in the human ocular
fundus measured by reflectance oximetry: preliminary data in retinal veins.
Klin Monatsbl Augenheilkd 2002 Apr;219(4):289-91.
34 Burgansky-Eliash
Z, Nelson DA, Bar-Tal OP, Lowenstein A, Grinvald A, Barak A. Reduced retinal
blood flow velocity in diabetic retinopathy. Retina. 2010 May;30(5):765-73.
PubMed PMID: 20061994.
35 Geirsdottir
A, Palsson O, Hardarson SH, et al. Retinal vessel oxygen saturation in healthy
individuals. Ophthalmol. Vis. Sci. 53: 5433-5442 (2009).
36 Hammer
M, Riemer T, Vilser W, et al. A new imaging technique for retinal vessel
oximetry: principles and first clinical results in patients with retinal
arterial occlusion and diabetic retinopathy. Proc. SPIE 7163: 1-3 (2000).
37 Hardarson
SH, Gottfredsdottir MS, Halldorsson GH, et al. Glaucoma filtration surgery and
retinal oxygen saturation. Invest. Ophthalmol. Vis. Sci. 50(11): 5247-5250 (2009).
Peer reviewer: Qianying Gao, Zhongshan Ophthalmic Center, Sun Yat-sen
University, 54 S. Xianlie Road Guangzhou 510060 People’s Republie of China.
Refbacks
- There are currently no refbacks.