Measuring oxygen saturation in retinal and choroidal circulations in rats using visible light optical coherence tomography angiography

Abstract

Visible light optical coherence tomography (vis-OCT) has demonstrated its capability of measuring vascular oxygen saturation (sO₂) in vivo. Enhanced by OCT angiography, the signal from microvasculature can be further isolated and directly used for sO₂ extraction. In this work, we extended the theoretical formulation for OCT angiography-based oximetry by incorporating the contribution from motion contrast enhancement. We presented a new method to eliminate the associated confounding variables due to blood flow. First, we performed in vitro experiments to verify our theory, showing a stable spectral derivative within the selected wavelength bands for sO₂ extraction. Then, we tested our method in vivo to measure retinal sO₂ in rats inhaling different gas mixtures: normal air, 5% CO₂, pure O₂, and 10% O₂. Absolute sO₂ values in major arterioles and venules in the retinal circulation can be accurately measured. In addition, we demonstrated the relative changes of sO₂ can be measured non-invasively from choriocapillaris immediately underneath the retinal pigmented epithelium (RPE) in rodents.

1. Introduction

Optical coherence tomography (OCT) demonstrated tremendous success in imaging three dimensional (3D) tissue anatomy with micrometer-level spatial resolution on a macroscopic length scale [1]. In the past two decades, the technique has been translated into various clinical practices [2]. In particular, OCT has joined the standard of care in ophthalmology, providing rich 3D anatomical images from, for example, various retinal layers, choroid, and optic nerve head. Besides the technical advances on high-speed and high-resolution retinal imaging, functional imaging of various hemodynamic parameters has emerged as an invaluable supplement to anatomical OCT imaging. Perhaps the most important developments of functional OCT in retinal imaging are the measurement of retinal blood flow [3–9], and label-free retinal OCT microangiography (OMAG) [10–12]. Because the neuroretina has a high metabolic demand, the blood flow in the retinal circulation can be an excellent indicator of the viability of the retina and has been associated to various retinal diseases [13–15].

Another important hemodynamic parameter of the retinal circulation is the hemoglobin oxygenation saturation or sO₂. The sO₂ difference between the arterial and venous blood vessels indicates oxygen extraction from the circulation. When combined with blood flow measurements, oxygen metabolism in the retina can be quantified by gas volume of oxygen consumed per unit time [16, 17]. The attempts of using OCT to measure blood oxygenation dated back more than a decade ago, where the spectroscopic contrast around the isosbestic point around 800 nm from oxy- and deoxy-hemoglobin was characterized [18–21]. Since then, several groups also explored the oxygenation dependent scattering coefficient to create contrast for oxygenated or deoxygenated whole blood. However, none of the efforts has progressed into successful measurements of absolute sO₂ in vivo. The reason is that the absorption contrast around 800 nm was overwhelmed by the wavelength-dependent scattering coefficient of red blood cells/whole blood, which have later been theoretically and numerically demonstrated [22, 23]. Knowing that hemoglobin has much higher absorbing contrast within visible-light spectral range than within NIR range, Yi et. al. and Robles et. al. explored the feasibility of using visible-light OCT (vis-OCT) to quantify sO₂ from ex vivo and in vitro samples, respectively [24, 25]. They further showed in vivo measurement of blood oxygenation [26], and the first demonstration of in vivo retinal oximetry in rodents using vis-OCT combing with the spatially-resolved spectroscopic analysis [27]. Later, to isolate the blood signal from the static tissue in OCT signal, we proposed to combine the speckle contrast and the spectroscopic contrast of blood simultaneously [28] to calculate sO₂ in microvasculature. We demonstrated that sO₂ could be directly calculated from wavelength-dependent OCT angiography without the cumbersome image processing previously needed to locate blood vessels. A similar strategy was recently adapted for both exposed cortical imaging and optical-power-cancelled retinal imaging in rodents [29]. However, the theoretical treatment did not consider the complication of blood flow to the angiography signal, which could confound the blood oxygenation calculation in vivo.

Here, we expanded our theoretical formulation by taking blood flow contrast into account in OCT angiography-based oximetry. Experimental verification was conducted by pumping anticoagulated whole bovine blood through a capillary tube. We tested the influence of different flow rates to the spectral derivative (i.e. the first-order partial derivative of spatially resolved OCT spectrum against wavelength). Further, we proposed a new data processing algorithm to normalize different flow rate and yield a robust sO₂ calculation. We tested our method in rat retina in vivo and detected sO₂ variation upon inhaling different gas mixtures. Moreover we demonstrated, for the first time, a non-invasive measurement of the relative sO₂ changes in choriocapillaris using vis-OCT.

2. Theory

In this section, we described the extended theoretical formulation for the spectroscopic analysis on vis-OCT angiography-based oximetry. The following motion-enhanced dynamic scattering model is to demonstrate that a linear relationship can be established between sO₂ and OCT angiographic derived spectrum measurements. Without loss of generality, we can start the formulation from a wavelength-dependent A-line signal from OCT angiogram $AA\left(z,\lambda \right)$, where z defines the axial coordinate and λ denotes the wavelength. Note that we will omit the finite axial resolution and sensitivity roll-off for simplicity purpose. At any given point along the A-line,

$$AA\left(z,\lambda \right)=I_0\left(\lambda \right)r\left(z,\lambda \right)\eta \left(z,\lambda \right)G\left(\Delta \phi \right),$$ (1)

where $I_0\left(\lambda \right)$ is incidence intensity. $r\left(\lambda \right)$ is the wavelength-dependent back scattering coefficient of local tissue, which can be further expressed by a power law $r\left(\lambda \right)=B{\lambda }^{-\alpha }$ under the first Born approximation. $B$ is a scaling constant while the value of $\alpha$ depends on the tissue type and has been previously characterized to be ~1 with variations around ± 0.2 [30, 31]. The term $G\left(\Delta \phi \right)$ is a non-negative scalar representing the dynamic contrast enhancement caused by random phase variation due to motion. The value of G is assumed to be 0 only if the tissue is static. $\eta \left(z,\lambda \right)$ is the cumulative attenuation when light double passes the tissue,

$$\eta (z,\lambda )=e^{-\underset{0}{\overset{z}{\int }}2{\mu }_t(z^{\text{'}}, \lambda )dz\text{'}},$$ (2)

where ${\mu }_t$ is the total attenuation coefficient of local tissue attributed to both optical absorption and scattering. For whole blood, it is a linear combination of absorption coefficient ${\mu }_a$, scattering coefficient ${\mu }_s$, and a packing factor W, ${\mu }_t={\mu }_a+W{\mu }_s$ [22, 32]. For other retinal tissue, reduced total attenuation coefficient, ${\mu }_t={\mu }_a+(1-g){\mu }_s$ may be used, where g denotes the anisotropy factor of the scattering process.

After normalizing $AA\left(z,\lambda \right)$ by the source spectrum $I_0\left(\lambda \right)$, the resulted image is analogous to the wavelength-dependent optical density (OD) mapping used in traditional fundus retinal oximetry, except for one additional motion enhancement term $G\left(\Delta \phi \right)$. Hence we use the same notation and denote the normalized $AA\left(z,\lambda \right)$ as $OD\left(z,\lambda \right)$,

$$OD\left(z,\lambda \right)=\ln \frac{AA\left(z,\lambda \right)}{I_0\left(\lambda \right)}=\ln r\left(\lambda \right)+\ln \eta \left({\mu }_a,{\mu }_s,\lambda \right)+\ln G\left(\Delta \phi \right),$$ (3)

or

$$OD\left(z,\lambda \right)=\alpha \lambda +\ln B-2\underset{0}{\overset{z}{\int }}{\mu }_t(z^{\text{'}},\lambda )dz\text{'}+\ln G\left(\Delta \phi \right).$$ (4)

As we take the partial derivative of OD against wavelength (spectral derivative), we have

$$\frac{\partial OD\left(z,\lambda \right)}{\partial \lambda }=-\frac{2\partial {\int }_0^z{\mu }_t(z^{\text{'}},\lambda )dz\text{'}}{\partial \lambda }+\alpha +\frac{\partial \ln G\left(\Delta \phi \right)}{\partial \lambda }.$$ (5)

If we assume that the wavelength is much smaller than the motion displacement of the blood cells, the phase variation becomes random and the enhancement factor $G\left(\Delta \phi \right)$ loses wavelength dependence. The spectral derivative of $G\left(\Delta \phi \right)$ could be considered to be zero and the equation becomes

$$\frac{\partial OD\left(z,\lambda \right)}{\partial \lambda }=-\frac{2\partial {\int }_0^z{\mu }_t(z^{\text{'}},\lambda )dz\text{'}}{\partial \lambda }+\alpha .$$ (6)

Now let us consider the attenuation contribution from the whole blood and the static tissues. As light propagates in tissue, the cumulative attenuation is

$$\underset{0}{\overset{z}{\int }}{\mu }_t(z^{\text{'}},\lambda )dz\text{'}=\underset{0}{\overset{l_b}{\int }}{\mu }_t(z^{\text{'}},\lambda )dz{\text{'}}_{blood}+\underset{0}{\overset{l_{else}}{\int }}{\mu }_t(z^{\text{'}},\lambda )dz{\text{'}}_{static tissue}$$ (7)

or

$$\underset{0}{\overset{z}{\int }}{\mu }_t(z^{\text{'}},\lambda )dz\text{'}={\mu }_{t\_b}{l}_b+\overline{{\mu }_{t\_else}}{l}_{else},$$ (8)

where ${\mu }_{t\_b}$ and $\overline{{\mu }_{t\_else}}$ denote the total attenuation coefficients of whole blood and averaged attenuation coefficient of static retinal tissue, respectively; $l_b$ and $l_{else}$ denote the cumulative optical path length within the blood vessel and in other retinal tissues, respectively.

Based on the previous characterization of ${\mu }_{t\_b}$ and $\overline{{\mu }_{t\_else}}$, we found that the spectral derivative of $\overline{{\mu }_{t\_else}}$ was close to zero [32, 33]. As a result, the spectral derivative of Eq. (7) is

$$\frac{\partial {\int }_0^z{\mu }_t(z^{\text{'}},\lambda )dz\text{'}}{\partial \lambda }=\frac{\partial {\mu }_{t\_b}}{\partial \lambda }{l}_b+\frac{\partial \overline{{\mu }_{t\_else}}}{\partial \lambda }{l}_{else}\cong \frac{\partial {\mu }_{t_b}}{\partial \lambda }{l}_b.$$ (9)

We can further express ${\mu }_{t\_b}$ with respect to sO₂ as

$${\mu }_{t\_b}=s{O}_2\cdot {\mu }_{t_{Oxy}}+\left(1-s{O}_2\right)\cdot {\mu }_{t_{Deoxy}},$$ (10)

where ${\mu }_{t\_Oxy}$ and ${\mu }_{t\_Deoxy}$ are the total attenuation coefficients for fully oxygenated and deoxygenated whole blood, respectively. For simplicity, we define the following two constants, $k_{Oxy}=\frac{\partial {\mu }_{t\_Oxy}}{\partial \lambda }$ and $k_{Deoxy}=\frac{\partial {\mu }_{t\_Deoxy}}{\partial \lambda }$ and plug Eqs. (8)-(10) into Eq. (6) so that

$$\frac{\partial OD\left(z,\lambda \right)}{\partial \lambda }=-s{O}_2\cdot \left(k_{Oxy}-k_{Deoxy}\right)l_b+k_{Deoxy}{l}_b+\alpha .$$ (11)

It is clear that Eq. (11) establishes a linear relationship between the spectral derivative of the OD function and sO₂, while $\left(k_{Oxy}-k_{Deoxy}\right)l_b$ serves as a constant scaling factor. During our experiment, l_b is a constant determined by the choice of ROI. Its exact value can be estimated by measuring the mean optical path on the OCT angiograph.

3. Methods

3.1 Animal preparation

We anesthetized wild-type Long Evans rats using Ketamine/Xylazine cocktail solution (11.45 mg Ketamine and 1.71 mg Xylazine per milliliter of solution, respectively). The solution was administrated via intraperitoneal injection (IP) at a dosage of 87 mg Ketamine and 13 mg Xylazine per kilogram of body weight. A supplementary dose (30% of the original volume) was administrated after 30 minutes of the initial injection to maintain the animal at deep anesthesia. We applied 0.5% Tetracaine hydrochloride ophthalmic solution for local eye anesthesia and 1% Tropicamide ophthalmic solution for pupil dilation. During the entire period of anesthesia, we maintained the core body temperature at 37 °C using an electronic heating pad with feedback control (FHC Inc). The animal was held on a custom-made imaging stage and properly immobilized. A pulse oximeter attached to the rear limb of the animal provided real-time monitoring of its heart rate and systemic arterial oxygen saturation.

The animal was first ventilated with normal air (21% O₂, 79% N₂) at 4.2 Standard Liter per Minute (SLPM). Then we supplied the following gas mixtures in the order of: (1) 5% carbon dioxide (21% O₂, 74% N₂, and 5% CO₂); (2) pure oxygen (100% O₂); and (3) 10% oxygen (10% O₂ and 90% N₂). For each gas inhalation, the animal was allowed to stabilize for at least 3 minutes before the measurement was taken. Additional time was allowed if the pulse oximeter reading showed large fluctuation (>2% deviation over 10 seconds). We supplied normal air for 3 minutes between changes. We took one additional measurement after the sequence with normal air ventilation. For each OCT measurement, the corresponding oximeter readings were also recorded simultaneously.

All the experimental procedures were approved by the Northwestern University IACUC and conformed to the Association for Research in Vision and Ophthalmology Statement on Animal Research.

3.2 Image acquisition

We developed a spectral-domain vis-OCT system to image rodent retina in vivo. A supercontinuum laser source (SuperK EXTREME EXW-6, NKT Photonics) provided broadband illumination. The output beam was delivered to the input arm of a free-space Michaelson interferometer via fiber optics. The beam splitter (CM1-BS013, Thorlabs) of the interferometer divided and redirected the beam to the reference arm and the sample arm. In the reference arm, a set of carefully chosen glass plates compensated for dispersions created by the unbalanced optical components. A continuously adjustable ND filter (NDC-50C-2M, Thorlabs) attenuated the beam intensity to enhance the dynamic range of the interference signal. The sample arm consists of a pair of scanning galvanometer mirror (Nutfield Technology) and a Keplerian telescope (0.2 × magnification) to project rectangular raster scanning pattern on the rat fundus. The spectrometer covered a spectrum range from 520 nm to 630 nm. We estimated the transverse and axial resolutions of our vis-OCT system to be 15 μm and 1.7 μm, respectively. A fast CMOS line scan camera (spL2048-140km, Basler) digitized the interference spectrum for image reconstruction. All acquisition sequence and recording were synchronized by a customized LabView (National Instrument) virtual instrument program.

We adopted uni-directional scans in our OCT angiography. Here we denote the fast-scan direction as the x axis and slow scan direction as the y axis. During one B-scan, we translated the laser spot along the positive x axis at 50-kHz A-line rate and a total of 400 A-lines were recorded. Immediately after each B-scan, we returned (flied back) the laser spot to the x axis origin without acquiring any signal. The returning speed was limited to four times the forward scanning speed, which was constrained by the mechanical properties of the galvanometer. We repeated the process five times at each y location before moving to the next one. For one vis-OCT angiography image stack, we had 5 × 512 = 2560 B-scans at 512 different y locations, each B-scan consisted of 400 A-lines. The entire OCT image stack took 25.6 seconds to acquire.

3.3 Vis-OCT imaging of retinal microvasculature

We generated vis-OCT angiography using phase-sensitive decorrelation algorithm [11, 28, 34]. The contrast originated from the flowing blood cells inside vessels as they randomly change the phase and amplitude of back-scattered light. The random phase shift caused decorrelation in OCT signal and can be enhanced via high-pass filtering. However, such enhancement may be confounded by bulk motion, which also contributed to the signal via the same mechanism. These motion originated from normal physiological movements (e.g. heart beat and breathing) and the rotating galvanometer mirrors, both of which were inevitable. We must take measures to minimize their effects so as not to interfere with vis-OCT angiography. We adopted a three-step approach to correct the aforementioned bulk motion decorrelation within each y location: (1) Bulk image shift; (2) Axial global phase fluctuations (AGPF) correction; and (3) Lateral global phase fluctuations (LGPF) correction [35]. Without loss of generality, we used the notation $A(x,z,t)$ to represent the recovered complex vis-OCT signal of the five consecutive B-scans at a given y location; x and z represents the lateral and axial location, respectively; t indicates the specific B-scan in time order, whose value ranges from 1 to 5 in our case.

The first step was to co-register the five real-valued OCT structure image at each y location, so that motion shift greater than the imaging resolution can be corrected. We calculated the cross-correlation function between the two adjacent B-scan images (t + 1 and t). We identified the maximum value of the cross-correlation function along x and z directions, which indicates when the two images were properly aligned. The second image was then shifted back so the two images were properly registered.

The second step was to correct the global phase fluctuations with two phase modifiers AGPF and LGPF on the co-registered OCT images. The AGPF can be estimated by integrating the phase variation along the A-line axial direction (z axis) as

$$AGFP\left(x\right)=-arg\left[\int A^{*}\left(x,z,t_2\right)A\left(x,z,t_1\right)dz\right],$$ (12)

while the LGPF can be estimated by integrating the phase variation along the x axis as

$$LGFP\left(z\right)=-arg\left[\int A^{*}\left(x,z,t_2\right)A\left(x,z,t_1\right)dx\right].$$ (13)

We multiplied the two phase modifiers AGFP and LGFP to the corresponding complex OCT signal pixel-wise. The corrected complex valued vis-OCT signal is denoted $A\text{'}\left(x,z,t\right)$. The B-scan angiography cross-section $AG\left(x,z,t\right)$ at location y can then be calculated by

$$AG\left(x,z\right)=\mathbb{E}\left|A\text{'}\left(x,z,t_{\tau +1}\right)-A\text{'}\left(x,z,t_{\tau }\right)\right|,$$ (14)

where $\mathbb{E}$ represents the expected value; and $|\cdot |$ represents the modulus operation. We iterated the process at each y location and generated a three dimensional image stack $AG\left(x,y,z\right)$.

At last, en face angiogram was generated by integrating the first 15 pixels of the highest motion contrast along z direction.

3.4 sO₂ estimation

We performed spectroscopic angiography to obtain the spectral profile of motion enhanced contrast of blood vessels [10, 11, 28], which was used to calculate the absolute blood sO₂. Instead of using the full spectrum, we employed short-time Fourier transform (STFT) to perform spatial-spectral dual-domain analysis. We empirically chose four Gaussian-shaped STFT windows centered at 565 nm, 568 nm, 571 nm, and 574 nm, each with 10 nm full-width-half-maximum (FWHM) bandwidth. The full spectrum was windowed using these narrow-band STFT windows, respectively. For each center wavelength chosen, we generated the corresponding narrow-band OCT angiography using the method described in the above section. The calculated wavelength-dependent OCT angiography is denoted $AG\left(x,y,z,\lambda \right)$. For any given vessel at depth z, we first performed image segmentation to isolate a thin layer containing the vessel. To project the image stack into a two-dimensional en face angiograph $AG\left(x,y,z,\lambda \right)$, we integrated the first 15 superficial pixels within the blood vessel. The integration interval of 15 was empirically chosen to optimize signal-to-noise ratio. We then calculated the mean value of $AG\left(x,y,z,\lambda \right)$ within the x-y plane inside the vessel region. The value was then assigned to $AA\left(z,\lambda \right)$ to characterize one designated vessel, as in Eq. (1). The process is repeated for all four center wavelength windows.

Next, we normalized $AA\left(z,\lambda \right)$ using the corresponding source spectrum I₀(λ) to calculate the localized $OD\left(z,\lambda \right)$. We performed a linear regression to calculate the slope of the straight line that went through the center of the four OD points. We used this slope to estimate the partial derivative of the OCT angiograph spectrum against wavelength.

To convert the spectral derivative into absolute sO₂ values, we performed a calibration based on pulse oximetry spO₂ readings where we assumed that the major retinal arteries had the same sO₂ as systemic arterial sO₂. Based on the calibration on each animal, we can measure the absolute sO₂ values in the major retinal arteries and venules around the optic disk.

We summarized the data processing steps in Table 1.

Table 1. Summary of data processing procedures to extract sO_2.

	Operations procedure to extract sO2 for any given blood vessel
1	To generate wavelength-dependent OCT angiography using STFT and motion enhanced dynamic scattering contrast according to Eq. (14)
	Input: Raw interference spectrum, STFT window; Output: $AG\left(x,y,z,\lambda \right)$
2	To choose a region of interest within a vessel from en face OCT image and average the signal over x, y extents.
	Input: $AG\left(x,y,z,\lambda \right)$, ROI of a vessel (x, y); Output: $AA\left(z,\lambda \right)$
3	To calculate localized optical attenuation function $OD\left(z,\lambda \right)$ according to Eq. (4)
	Input: $AA\left(z,\lambda \right)$, source spectrum I0(λ); Output: $OD\left(z,\lambda \right)$
4	To calculate partial derivative on $OD\left(z,\lambda \right)$ against wavelength according to Eq. (5)
	Input: $OD\left(z,\lambda \right)$, λ; Output:$\frac{\partial OD\left(z,\lambda \right)}{\partial \lambda }$ (spectral derivative)
5	To calculate average $\frac{\partial OD\left(z,\lambda \right)}{\partial \lambda }$ long the z axis and to calibrate all the major retinal arteries against systemic spO2
	Input: spO2, $\frac{\partial OD\left(\lambda \right)}{\partial \lambda }$; Output: calibration curve
6	To convert $\frac{\partial OD\left(\lambda \right)}{\partial \lambda }$ to sO2 according to Eq. (11)
	Input: $\frac{\partial OD\left(\lambda \right)}{\partial \lambda }$ of selected vessel, calibration curve; Output: absolute sO2

4. Results

4.1 Phantom experiment verification

To demonstrate that the spectral derivative method is not affected by flow velocity, we designed and performed the following in vitro experiment. We fabricated a capillary phantom using a polystyrene tube (inner diameter: 125 µm) filled with anticoagulated bovine whole blood (Quad Five). We fully oxygenated the blood by exposing it to normal air at room temperature for 1 hour prior to the experiment. The phantom was embedded beneath 2.5% agarose gel to mimic a blood vessel in the retina. We controlled the blood flow rate in the tube using a syringe pump (PHD 2000 Infusion, Harvard Apparatus) at various pre-set values within physiological range.

In order to image the in vitro sample, we slightly modified our OCT system. An objective lens (LSM03-VIS, Thorlabs) replaced the telescope system to create a focusing spot about 5.5 µm in diameter. The same data acquisition and processing protocols were applied.

Figure 1(a) compares the cross-sectional view of OCT structural images and motion-enhanced images. It can be clearly seen that surrounding agarose gel has strong scattering background in conventional OCT. In comparison, the motion-enhanced images largely suppressed the static background, while the moving blood within the tube was enhanced due to dynamic scattering. Even when the flow rate was set to zero, intrinsic random Brownian motion was sufficient to enhance the contrast (the left most images in Fig. 1(a)). On the sectioned en face view of the OCT angiograph showing the capillary phantom, we took four rectangular region of interest (ROI’s) of equal size along its longitudinal axis (Fig. 1(b)). We extracted the wavelength-dependent motion contrast spectrum within the selected ROI’s to evaluate the influence of flow velocity. As expected, the shapes of these spectra were comparable, showing characteristic absorption spectrum of oxygenated whole blood. The spectrum calculated from zero-flow phantom is shown in Fig. 1(c).

Fig. 1.

Spectroscopic analysis of in vitro phantom experiment. (a) Cross-sectional images of the tube based on direct (top row) and motion-enhanced dynamic scattering (bottom row) contrasts; (b) En face OCT angiography image of the tube. The yellow rectangles indicate the ROI used; (c) Representative motion contrast spectrum calculated from selected ROI’s, flow rate = 0.00 mL/min; (d) Averaged intensity I_m taken from four ROI’s within the capillary tube; (d) The spectral derivative of motion contrast spectrum obtained within the capillary tube with varying flow rates from 560 to 580 nm. Error bar = S.D.

We extracted two parameters to characterize the obtained motion contrast spectrum, the averaged intensity (I_m) and spectral derivative (∂I/∂λ) from 565 nm to 574 nm. Their values were plotted against increasing flow rate in Figs. 1(d) and 1(e), respectively. We performed liner regressions to estimate the slopes of both curves. I_m showed a clear decreasing trend with a negative slope, which is statistically significant (p < 0.001). On the contrary, the slope of the ∂I/∂λ curve did not deviate from zero (p = 0.74). The result showed that ∂I/∂λ was flow-independent across the tested range, suggesting that sO₂ estimation based on ∂I/∂λ would not be affected by different flow rate.

4.2 Measuring rat retinal sO₂ in vivo

Figure 2 compares the cross-section view of the OCT images in rat retina in vivo. Figure 2(a) is the conventional structural OCT image. The anatomical features of the retina can be clearly distinguished. The cross sectional OCT angiogram is shown in Fig. 2(b). We further overlaid the structure image with the angiography in Fig. 2(c), showing locations of each vessel within the retina. Besides the vasculatures in retinal circulation, we also found strong blood motion contrast directly beneath the RPE, which suggested the existence of blood flows in choriocapillaris. Beyond that, visible light attenuated rapidly.

Fig. 2.

(a) Example of cross-sectional images of conventional OCT in rat retina in vivo using vis-OCT. Anatomical layers are labeled including, NFL/GCL-nerve fiber layer/ganglion cell layer, IPL-inner plexiform layer, INL-inner nuclear layer, OPL-outer plexiform layer, ONL-outer nuclear layer, ELM-external limiting membrane, IS/OS-inner segments/outer segments, and RPE-retinal pigment epithelium. (b) The corresponding OCT microangiograph generated using motion contrast. Vessels in retinal circulation and choroicapillaries are identified. Artifacts created by the hyper-reflective layers in IS/OS and RPE are indicated with a blue arrow. (c) Overlaid image from (a) and (b). The choroicapillaris lay beneath RPE layer.

We then compared the en face view of rat fundus obtained by two OCT modality. Figure 3(a) illustrates the conventional structural OCT image. Although the superficial vessels can be clearly observed, we cannot distinguish the microvasculature. In comparison, Fig. 3(b) showed motion-enhanced dynamic scattering contrast of the same data set, where the hyper-reflective signals from the static tissues are largely suppressed. Due to limited lateral resolution, choriocapillaris appeared as a single bright layer and was excluded through segmentation. Further, we calculated the sO₂ of the major retinal vessels in the retinal circulation during normal oxygen inhalation, as the color-encoded angiographic image in Fig. 3(b). We can clearly distinguish the arteries and veins based on their individual sO₂. Finally, we rendered the 3D vascular structure in Fig. 3(c). Besides the dominating brunches of central retinal arteries and veins, laminated structures of micro-vasculatures within nerve fiber layer (NFL) and outer plexiform layer (OPL) are also distinguishable. The fly-through en face microangiogram is shown in the movie.

Fig. 3.

In vivo imaging of a rat retina using vis-OCT. (a) and (b) are en face projection of amplitude and motion enhanced microangiography, respectively. In the microangiography map, each major retinal arterioles and venules are color-encoded based on the measured sO₂ value. (c) A three dimensional rendering of rat retinal microvasculature. See Visualization 1 ^{(4.8MB, AVI)} for the fly-through en face microangiogram.

4.3 Absolute sO₂ measurement in retinal circulation

We next performed in vivo verification of our spectral derivative method. We calibrated the absolute sO₂ values from the major retinal arteries against a pulse oximeter readings under different ventilation conditions (normal air, 5% CO₂, pure O₂, and 10% O₂), as shown in Fig. 4(a). The calibration process corresponded to the Step 5 in Table 1. The result shows a strong positive correlation (R² = 0.87) between the two independent measurements, indicating that the calibration offered reasonably good estimation of the true sO₂.

Fig. 4.

Measurement of absolute sO₂ values in the retinal circulation in vivo. (a) Measured arterial sO₂ calibrated against pulse oximetry readings (R² = 0.87) from n = 8 rats. Black line is the linear regression of the data (b) Measured sO₂ of major retinal arteries and venues upon changes of inhalation gas mixtures (n = 8 for each bar). Error bar = S.E.M. *: p < 0.05. **: p < 0.01.

Based on the established calibration curve, we measured the major retinal arterial and venous sO₂ in respond to the change of inhalation gases. In the course of the experiment, we changed the inhalation gases in the order of normal air, 5% CO₂, pure oxygen, and 10% oxygen. As shown in Fig. 4(b), the arterial and venous sO₂ were significantly different (p < 0.05), except when pure oxygen was inhaled and the shunting between arterioles and venules elevated the venous sO₂. Besides the arteriovenous sO₂ difference, we observed a moderate increase (p = 0.06) in both arterial and venous sO₂ when we supplied 5% CO₂ gas. This increased sO₂ is expected because 5% CO₂ has the effect of vasodilation, which improves tissue oxygenation [36, 37]. Both the arterial and venous sO₂ increased significantly (p < 0.05) under pure O₂ and decreased significantly (p<0.01) under 10% O₂.

4.4 Relative sO₂ change in choriocapillaris

Given the fact that we had strong flow-contrast enhanced signals from choriocapillaris directly underneath the RPE, we attempted to measure the relative changes of sO₂ in choriocapillaris. Due to limited lateral resolution, we could not distinguish individual choriocapillaris; instead, we averaged the signal over the entire area of choriocapillaris. We performed image segmentation to remove any area shadowed by superficial blood vessels, and yielded one averaged spectrum $AA\left(z_c,\lambda \right)$, where z_c was the axial position in the choroid.

In order to eliminate the contamination from the retinal circulation in the inner retina, we normalized the choriocapillaris OCT angiograph spectrum by the intensity spectrum from the layer directly above RPE and calculated the spectral derivative. We used the same calibration curve obtained from retinal circulation to obtain the relative change of sO₂ with respect to normal air breathing, as in Fig. 5. When we supplied 5% CO₂ gas mixture, the average sO₂ in choriocapillaris was observed to increase moderately by 2% (p = 0.29). When we supplied pure O₂ and 10% O₂ gas mixture, the average sO₂ was observed to increase by 3.5% (p < 0.001) and decreased by 5% (p < 0.001), respectively.

Fig. 5.

Measured relative choriocapillaris sO₂ change comparing to normal air breathing (n = 6). ***: p < 0.0001.

4. Discussion

In this work, we presented an extended theoretical formulation for OCT angiography-based oximetry. This method used the flow contrast in OCT angiography, based on which we normalized the wavelength-dependent angiography and provided a robust measurement of sO₂. The approach differs from our previous vis-OCT oximetry publication [27], where only static OCT amplitude information was used. The introduction of dynamic contrast eliminates the need for cumbersome vessel segmentation, and has arguably higher sensitivity and noise resistance than our previous method. More importantly, the new approach also enables us to measure the relative change of sO₂ in chororiocapillaries, which are generally inaccessible in structural vis-OCT images. We observed the change of sO₂ in both retinal circulation and choriocapillaris under different inhaling gas mixtures, which include normal air, 5% CO₂, 100% O₂, and 10% O₂. Our in vivo measurements were verified by independent and concurrent pulse oximetry.

In our dynamic scattering model, we assumed that the motion enhancement term (G) is wavelength-independent. The assumption requires that the phase fluctuation is random between 0 to 2π without any preference. The condition is automatically satisfied when the displacement of scattering particles during the scanning dwell time are much larger than the wavelength. In our case, the red blood cell travels about 0.13 mm during the 0.01 s dwell time, which suffices the requirement.

We calculated the first derivative of the spectral range from 560 nm to 580 nm to establish its correlation with sO₂. It differs from the least square fitting method over much larger wavelength range (530 nm-600 nm) as previously reported by our and other groups [28, 29]. The major benefit is that the permitted incidence power in the current study could be reduced by about 2/3, which is ~0.3mW in the current experiments. By removing unnecessary exposure, the safety margin for vis-OCT for human can be significantly improved and the discomfort can be dramatically reduced. By sacrificing the spectral information from the absorption contrast, our method requires independent oxygenation measurements for calibration purpose. Fortunately, pulse oximetry utilizes the pulsatile profile and provides the reading of the overall arterial oxygen saturation, which should be equal to the values in the major retinal arterioles immediately exiting optic nerve head. This allows us to calibrate our spectral derivatives as we showed in Fig. 4(a) in every experiment. Such a calibration is easily achievable using a pulse oximetry that is widely available in the clinical settings.

Interestingly, we were also able to measure sO₂ variations in choriocapillaris right beneath the RPE under different inhalation conditions. The choroidal circulation is very different from retinal circulation, both anatomically and functionally [38]. The choroidal blood is supplied through posterior ciliary arterioles and drained through vortex vein. To our concern, the major difference between the two circulations is that the choroidal circulation has very small arteriovenous sO₂ difference, whereas the retinal circulation has a large arteriovenous sO₂ difference [39, 40]. This results in an almost uniform sO₂ level in the choriocapillaris under normal physiology conditions, which allows us to measure sO₂ over a relatively large area of choriocapillaris.

However, there are a few complications that prevent us providing absolute sO₂ measurements in choriocapillaris. First, the true sO₂ level in choriocapillaris is still unknown. The most accurate reference is the direct measurement of oxygen tension using an invasive oxygen sensitive microelectrode [38]. Although it is commonly assumed that the choroidal circulation has the systemic oxygenation as in the arterioles, the microelectrode measurements turned out to be lower than expected [38]. Without a reliable reference, we were not able to calibrate our spectral derivative as we did for the retinal vessels. Alternatively, one might consider using the least square fitting method with the full spectral range of the hemoglobin absorption contrast to characterize choriocapillaris sO₂. However, the optical properties of dispersed single blood cells in capillary network are not entirely clear, which would bring another uncertainty in the least square fitting method. Second, light passes through entire retina before choriocapillaris. The accumulated attenuation is also wavelength dependent, which would affect the spectral derivative. Nonetheless, we found that this is the first time a non-invasive measurement can be implemented to measure relative choriocapillaris sO₂ variations.

5. Conclusion

In summary, we presented a new theoretical formulation for vis-OCT angiography-based oximetry. The flow contrast was used and characterized by in vitro experiments. After proper spectral normalization, we demonstrated, for the first time, robust in vivo measurements of absolute sO₂ variation in the retinal circulation and relative sO₂ variation in the choriocapillaris. The new method requires much reduced illumination power, which greatly benefits clinical applications of vis-OCT.