2.1.

Surgical Procedure/Animal Preparation

The swine was administered a general anesthesia mixture consisting of air, nitrogen, oxygen, and isofluorane. A Swan-Ganz catheter was placed in the aorta just distal to the heart, and the second catheter was placed in the inferior vena cava just proximal to the heart. Blood samples collected from the two Swan-Ganz catheters were analyzed with an ISTAT Personal Clinical Analyzer (Heska Corporation, Fribourg, Switzerland) to monitor systemic arterial oxygen saturation $\left({\mathrm{S}}_{\mathrm{a}}{\mathrm{O}}_2\right)$ and mixed venous oxygen saturation $\left({\mathrm{S}}_{\mathrm{v}}{\mathrm{O}}_2\right)$ . The $\mathrm{S}{\mathrm{O}}_2$ , $\mathrm{P}{\mathrm{O}}_2$ , $\mathrm{P}\mathrm{C}{\mathrm{O}}_2$ , pH, and hematocrit were measured from the small blood samples. A peripheral oxygen saturation monitor was also placed in the rectum of the swine. A lid speculum was placed in the operative eye. The swine was inspired with 100% oxygen throughout the data collection.

A $4\text{-}\mathrm{mm}$ infusion cannulum was placed through the pars plana and sutured in place. The infusion cannulum was attached to a balanced salt solution gravity feed system to maintain intraocular pressure. A contact vitrectomy lens (F36202.08, Bausch & Lomb, Rochester, New York) was placed onto the cornea using an index matching (refractive index: $n_d=1.337$ ) viscoelastic coupling agent.

2.2.

Fundus Imaging Apparatus and Data Collection

Following the surgical preparation of the animal, light from a scanning monochromator (Oriel Spectral Luminator, Irvine, California) was coupled into a fiber optic intravitreal illuminator (Alcon Laboratories, Fort Worth, Texas) that was inserted through the pars plana into the vitreous to illuminate the retina. The spectral resolution of the monochromator was approximately $10\mathrm{nm}$ . A $12\text{-}\text{bit}$ scientific grade charge-coupled device (CCD) camera (Hamamatsu Orca-AG, Hamamatsu, Japan) with a $3.3\times$ macro zoom lens was used to image the fundus. The peak wavelength of the $10\text{-}\mathrm{nm}$ wavelength band was stepped from $420\text{to}700\mathrm{nm}$ , and an image of the target area was acquired at illumination center wavelengths of 420, 430, 440, 460, 480, 490, 500, 510, 521, 532, 540, 545, 548, 552, 555, 558, 561, 565, 570, 575, 580, 590, 600, 615, 630, 645, 660, 680, and $700\mathrm{nm}$ . Each camera exposure was triggered using a cardiac monitor (Digicare Life Windows, Boynton Beach, Florida) to minimize fundus thickness variations by acquiring all images at the same point during the cardiac cycle; retinal vessel diameter variations can range from 2% to 17% due to the cardiac cycle.^{21, 22, 23} Additionally, the exposure time was adjusted to less than $40\mathrm{ms}$ (approximately $1∕25$ of the cardiac cycle) so the retinal blood vessel size should not vary appreciably during the exposure. The optical system is illustrated in Fig. 1 .

Fig. 1

Intravitreal illumination method for fundus reflectance measurements.

Multispectral image sets were obtained over a $1.39\mathrm{mm}\times 1.39\mathrm{mm}$ region of the illuminated fundus of the intact eye. A dark image (obtained with the illuminator shutter closed) was subtracted from each image upon acquisition to correct for dark current and external light sources. The reflected light was collected over a solid angle of 0.028 sr, the solid angle subtended by the eye’s pupil, which is the aperture stop of this system. After obtaining the spectral image set for the intact eye, the vitreous cutter was placed in the eye and a complete pars plana vitrectomy was performed.

After the postvitrectomy data was collected, a scleratomy was created to allow the insertion of a small disk $(3\mathrm{mm}\times 1\mathrm{mm})$ of Spectralon. The Spectralon was placed on top of the fundus area previously imaged and spectral images were obtained. A retinal detachment was induced near the site of the postvitrectomy measurements and the Spectralon disk was inserted underneath the retina and the transmittance of the isolated retina was measured for the same field of view. Figure 2 shows a drawing of the placement of the subretinal Spectralon chip.

Fig. 2

A retinal detachment was induced using perfluoron and a small, thin Spectralon disk was carefully inserted between the retina and RPE.

2.3.

Data Calibration and Image Alignment

A calibration data set was collected by measuring the output port of an integrating sphere illuminated with the fiber optic illuminator. The multispectral images were corrected for spectral variation in the light source and the spectral response of the CCD by dividing by the calibration spectrum. The inner wall of the integrating sphere is coated with Spectralon, which has a very uniform spectral reflectivity and is commonly used as a calibration standard.²⁴

The fundus multispectral image sets were registered to correct for any relative motion between the swine fundus and the imaging camera that occurred during data collection. Images were aligned to within a fraction of a pixel using a bicubic spline interpolation in conjunction with a maximum mutual information merit function algorithm.²⁵

2.4.

Image Processing and Data Analysis Techniques

Three sites relatively free of large vessels were chosen for analysis of fundus reflectance as indicated by the squares in Fig. 3 . Multispectral image sets were acquired for the following four experimental conditions: (1) from the intact eye, (2) from the eye postvitrectomy, (3) after insertion of subretinal Spectralon, and (4) for super-retinal Spectralon. The spectrum was calculated for each condition by averaging the flux over a small region $(100\times 100\text{microns})$ of the fundus shown in Fig. 3 for each of the 29 monochromatic images.

Fig. 3

The $523\text{-}\mathrm{nm}$ image for each of the 4 experimental conditions is shown. Reflectance spectra were averaged over each of the squares.

The signal recorded by the CCD camera is a product of the spectral irradiance of the light source, the spectral response of the imaging system, and the spectral transmittance or reflectance of any structures that may interact with the light captured by the imaging system. This relationship is expressed in equation form for each of the four experimental conditions described in the previous paragraph (denoted $S_{\mathit{IE}}$ , $S_{\mathit{PE}}$ , $S_{\mathit{sub}}$ , and $S_{\mathit{\text{super}}}$ , the signal recorded for experimental conditions 1 through 4, respectively) as follows:

2.4.1.

Low pass filtering and light/fundus relative motion correction

Despite attempts to immobilize the swine’s head during data collection, small relative motion between the illumination beam and the fundus occurred as a result of the swine’s breathing. This motion is in addition to the relative motion between the fundus and the camera, which was corrected for using the maximization of the mutual information as described in Sec. 2.3. Low pass filtered images (Fig. 4 ) reveal the relative motion between the illumination and the swine fundus; although the physical structures of the fundus were aligned for all the spectral images, the location of the peak illumination region varied among the spectral images. Figure 4 shows four low pass filtered images (420, 523, 560, and $680\mathrm{nm}$ ) obtained for the intact eye.

Fig. 4

Four low pass filtered images of the intact eye. The small cross on each image marks the location of the maximum intensity, corresponding to the axis of the fiber optic illuminator, or the peak intensity of the incident cone of light. The location of the maximum intensity moves as the swine breathes.

To analyze the motion of the illuminating beam, frequency domain filtering with a Gaussian low pass filter was performed on each of the monochromatic images. The filter, $H(\xi ,\eta )$ , is a Gaussian low pass filter with a spatial domain kernel size of $56\text{pixels}$ for a $102\times 102\text{pixel}$ image (full width at half $\text{maximum}=56\text{pixels}$ ),

Eq. 5

$$H(\xi ,\eta )=\exp -\left[\frac{{(\xi -\frac{N}{2})}^2+{(\eta -\frac{N}{2})}^2}{{\left(2\sigma \right)}^2}\right].$$

Given the Fourier transform of the image $i(\xi ,\eta )$

Eq. 6

$$i(\xi ,\eta )=\frac{1}{N^2}\sum _{x=1}^{N-1}\sum _{y=1}^{N-1}I{(x,y)}^{*}\exp [-i2\pi (\frac{\xi x}{N}+\frac{\eta y}{N})],$$

the low pass filtered image, $I_{\mathit{\text{filtered}}}Ix,y$ ),

Eq. 7

$$I_{\mathit{\text{filtered}}}(x,y)=F^{-1}\left\{H{(\xi ,\eta )}^{*}i(\xi ,\eta )\right\},$$

provides the centroid of the illumination as shown in Fig. 4. Figure 5 shows the frequency [Fig. 5a] and spatial domain [Fig. 5b] representation of the filter $H(\xi ,\eta )$ .

Fig. 5

Low pass filter in the (a) frequency and (b) spatial domains.

The beam centroid motion as a function of illumination wavelength is determined from the filtered images. This small relative motion causes irradiance fluctuations leading to errors in the measured spectrum. An algorithm was developed to correct for the motion of the illuminating irradiance: (1) for the filtered reference image (the $552\text{-}\mathrm{nm}$ image), the ratio between the maximum of the image and the value at the analysis site is calculated; (2) the corresponding ratios are determined for all the filtered images in the spectral image set; and (3) a correction factor, $c.f.=r_{\mathit{\text{base}}}∕r_{\mathit{\text{test}}}$ , is applied to the reflectance measured at the illumination site for each spectral image. The assumption is that the illumination profile does not change appreciably over the visible spectrum. Figure 6 illustrates the motion correction for one of the reflectance curves.

Fig. 6

An example of the effect of illumination motion correction. The solid line is the signal recorded by the CCD, and the dashed line shows motion-corrected spectra.

2.4.3.

Power compensation: estimation of absolute reflectance

The constant term $P_i$ in Eqs. 1, 2, 3, 4 was calculated to provide the absolute fundus reflectance and sensory retina transmittance. A calculation of $P_i$ is described for the intact eye; the other three conditions were similarly analyzed.

By taking the logarithm of Eq. 1, a set of linear equations for all ${\lambda }_i$ is formed

Eq. 12

$${\log }_{10}\left[\frac{S_{IE}\left(\lambda \right)}{\Theta {\left(\lambda \right)}^{*}{T}_{OM}\left(\lambda \right)}\right]={\log }_{10}\left[P_{IE}\right]+{\log }_{10}\left[R_{\mathit{\text{fundus}}}\left(\lambda \right)\right],$$

where all known terms are grouped on the left-hand side and $T_{\mathit{OM}}\left(\lambda \right)$ is the transmittance of the ocular media. An appropriate model of $R_{\mathit{\text{fundus}}}\left(\lambda \right)$ provides an overdetermined set of linear equations used to solve for $P_{\mathit{IE}}$ . $R_{\mathit{\text{fundus}}}^{\prime }\left(\lambda \right)$ , $T_{\mathit{SR}}\left(\lambda \right)$ , and $R_{\mathit{\text{fundus}}}^{″}\left(\lambda \right)$ are similarly modeled and the resulting linear equations solved for $P_i$ . A fundus reflectance model, Eq. 13, is used to model the measured quantity for all four experimental conditions based on the assumption that a finite number of reflecting, transmitting, and absorbing layers with known spectral extinction coefficients shape the spectral profile of light reflecting from the fundus. The primary ocular constituents that contribute to the spectrum are the ocular media (lens, cornea), hemoglobin, and melanin, although several other absorbers in smaller concentrations are likely to be present.^{5, 26} The reflectance model is

Eq. 13

$$R_{\mathit{\text{fundus}}}\left(\lambda \right)=T_{OM}{\left(\lambda \right)}^{*}{A}_{Hb}{\left(\lambda \right)}^{*}{A}_M{\left(\lambda \right)}^{*}{A}_{MP}\left(\lambda \right),$$

where $T_{\mathit{OM}}\left(\lambda \right)$ is the transmittance of the ocular media, $A_{Hb}\left(\lambda \right)$ is absorption due to blood (and is dependent on the oxygen saturation), $A_M\left(\lambda \right)$ is absorption due to melanin, and $A_{\mathit{MP}}\left(\lambda \right)$ is the absorption due to macular pigments. Figure 7 illustrates the absorption curves.

Fig. 7

Optical density of the dominant spectrally absorbing constituents of the eye (Refs. 26, 31).

Examination of Fig. 7 shows that for the wavelength range of $575\text{to}700\mathrm{nm}$ , the only components with substantial absorption are melanin and blood. Using the fundus reflectance model in the aforementioned wavelength range, Eq. 12 can then be rewritten to include the fundus reflectance model as

14.

$${\widehat{D}}_{IE}\left(\lambda \right)={\widehat{P}}_{IE}+{\widehat{A}}_{Hb}\left(\lambda \right)+{\widehat{A}}_M\left(\lambda \right),$$

$${\widehat{A}}_{Hb}\left(\lambda \right)={\epsilon }_{Hb}\left(\lambda \right)*c*d,$$

$${\widehat{A}}_M\left(\lambda \right)=M\left(\lambda \right)*Y,$$

where ${\widehat{D}}_{\mathit{IE}}\left(\lambda \right)$ represents the left-hand side of Eq. 12. The optical density due to blood is ${\widehat{A}}_{Hb}$ (the “̂” symbol denotes base 10 logarithm) and is described by the product of the millimolar extinction coefficient $\left({\epsilon }_{Hb}\right)$ , the concentration of hemoglobin $\left(c\right)$ , and the effective interaction path length $\left(d\right)$ . The oxygen saturation of the blood in each of the four cases was calculated using the method described by Denninghoff ²⁸ to be 85%, 52%, 92%, and 92% for the four cases and the corresponding millimolar extinction coefficients were used in Eq. 14. ${\widehat{A}}_M$ is the optical density due to melanin (found mainly in the RPE), which is a product of $M$ , the millimolar extinction coefficient and the relative concentration of melanin $\left(Y\right)$ . Writing Eq. 14 for the illumination wavelengths from $575\text{to}700\mathrm{nm}$ provides a set of overdetermined linear equations that was solved using the pseudo-inverse technique to provide the least-squares solution for $P_i$ .

3.1.

Diffuse Relative Reflectance Measurements

Figure 8 shows the measured relative fundus reflectance for the four experimental conditions. Figures 8a and 8b show similar behavior for the intact eye and the postvitrectomy data: reflectance is lowest in the blue light and increases until $510\mathrm{nm}$ for the intact eye and $480\mathrm{nm}$ for postvitrectomy. Oxyhemoglobin is evident by the appearance of bimodal local minima at 542 and $576\mathrm{nm}$ ; these minima correspond to the absorption maxima of oxyhemoglobin. The reflectance increases by a factor of approximately 6 for all three curves in each plot in the $420\text{-}\text{to}700\text{-}\mathrm{nm}$ range, and by a factor of approximately 2 for the $590\text{-}\text{to}700\text{-}\mathrm{nm}$ range. Figures 8c and 8d show markedly different behavior, although some similarities exist. For both the sub- and super-retinal Spectralon, a local maximum exists at $510\mathrm{nm}$ , and again, the oxyhemoglobin absorption spectrum is evident. The reflectance in this case, however, increases by a factor of approximately 4 in the $420–\text{to}700\text{-}\mathrm{nm}$ range, and a factor of approximately 1.75 in the $590–\text{to}700\text{-}\mathrm{nm}$ range.

Fig. 8

Relative reflectance measurements for the (a) intact eye, (b) postvitrectomy, (c) subretinal Spectralon, and (d) super-retinal Spectralon.

3.2.

Absolute Diffuse Relative Reflectance Measurements

Using the technique described in Sec. 2, the relative reflectance data was converted to absolute reflectance by solving for the $P_i$ coefficients from Eqs. 8 to 11. The results for the reflectance curves measured at the three analysis sites are included in Table 1 , and the power corrected curves are shown in Fig. 9 .

Fig. 9

Absolute reflectance measurements at the three analysis locations [(a) to (c)] after correction for motion and power of the illuminating beam. In each plot, the separate curves correspond to the four experimental conditions. The super-retinal spectrum contains two components: (1) light that reflects from the Spectralon and (2) light that has also scattered from the globe and into the imaging system. The latter contribution contains a strong hemoglobin signature.

Table 1

Amounts of hemoglobin and melanin absorption and the Pi parameter for the four conditions and three locations. The c*d product gives a measure of the amount of blood at a test site, the Y term indicates the amount of melanin, and the Pi parameter is proportional to the incident flux at each analysis site.

The relative melanin $Y$ values calculated for the intact eye and the postvitrectomy eye are much larger than the corresponding Spectralon values for all three locations in the data. This behavior is to be expected: The light striking both the subretinal and super-retinal Spectralon should see only a small fraction of the melanin encountered for the first two conditions. In fact, the only optical path for melanin absorption in the super-retinal Spectralon data is scattered light from the globe. The values obtained for the oxyhemoglobin $c^{*}d$ products correspond to a blood thickness interaction of $40\text{to}100\mu \mathrm{m}$ .

The absolute reflectance data shown in Fig. 9 is arranged such that the stacked plots in the left-hand column display the data from the three analysis sites. For all three locations, the reflectance for the first two cases is similar (intact eye and postvitrectomy), with the primary difference between the two curves being a difference in oxygen saturation (measured at 85% and 52%, respectively). The reflectance is approximately 17% at $420\mathrm{nm}$ , rises to local maxima of approximately 25% at 497 and $485\mathrm{nm}$ for the intact eye and postvitrectomy eye, respectively. From there, both curves decrease to bimodal local minima at 542 and $576\mathrm{nm}$ , and then increase monotonically from a value of $\approx 30\%$ at $590\mathrm{nm}$ to $\approx 60\%$ at $700\mathrm{nm}$ .

The Fig. 9 plots for subretinal Spectralon show more variation, ranging from a reflectance of $\approx 19\%$ to 25% at $420\mathrm{nm}$ . Above this wavelength, the reflectance increases to a local maximum of $\approx 48\%$ to 62% at $500\mathrm{nm}$ . The characteristic bimodal maxima are located at 542 and $576\mathrm{nm}$ . The reflectance again increases from the $575\text{-}\text{to}700\text{-}\mathrm{nm}$ range by 55% to 62% for $575\mathrm{nm}$ to 75% to 80% for $700\mathrm{nm}$ . The super-retinal Spectralon spectrum contains considerably more noise than the other spectra, possibly due to less exposure time in response to high reflectivity. The super-retinal Spectralon should be the highest recorded reflectance spectra in all three cases of Fig. 9, but in Fig. 9b was corrected to a value lower than that of the subretinal Spectralon data. This is likely due to the failure of our power fluctuation compensation method to completely match the power in the region and noise combined with the very high transmission of the sensory retina. In most cases [Figs. 9a, 9b, 9c], however, the super-retinal Spectralon had a higher reflectance than the subretinal Spectralon.

The transmittance of the sensory retina in a single pass can be obtained by taking the square root of the subretinal Spectralon data in Fig. 9, which were acquired in double passes. Figure 10 represents the transmittance of the sensory retina averaged over all three locations based on the power-corrected Eq. 10. The spectrum overestimates the influence of both melanin and hemoglobin due to scattered light from the globe. Equation 11 was used to calibrate for this scattered light term. The super-retinal spectrum calibrates the transmission through the anterior chamber and the scattered light from the globe itself. The latter component contains a significant hemoglobin signature. By dividing Eq. 10 by Eq. 11, the scatter corrected in vivo transmittance of the sensory retina is obtained (Fig. 11 ),

Fig. 10

Single pass transmittance of the sensory retina, averaged over the three locations measured from the subretinal Spectralon. The illumination component from light scattered from the globe acquired a strong hemoglobin signature, which overestimated the amount of blood contained in the sensory transmittance.

Fig. 11

Single pass transmittance of the sensory retina, averaged over the three locations measured from the subretinal Spectralon data and corrected for the influence of scattered light using the super-retinal data.

Figure 12 displays the expected transmittance of the sensory retina, which is nearly flat across the visible spectrum with an absolute transmittance $>90\%$ for all wavelengths.

Fig. 12

The measurement area is illuminated by two mechanisms: (1) illumination proved directly from the fiber optic illuminator and (2) scattered light from the globe that behaves like an integrating sphere illuminator.