Since the inception of optical coherence tomography (OCT) in the early 1990s [1,2], OCT research has focused predominantly on intrinsic optical contrast sources, such as reflectivity and motion. Tissue properties such as absorption spectra and birefringence have also been investigated for further classification of biological structures. Each of these can be measured in vivo without the need for exogenous contrast agents, unlike other optical techniques such as fluorescence angiography (FA). More recently, however, there has been growing interest in the application of exogenous contrast agents to enhance OCT signals and to provide additional biological information, such as tracer kinetics [3] or quantitative blood volume and flow [4]. Some of the contrast agents that have been applied to OCT include Intralipid [3–5], which is an emulsion of lipid particles, gold nanoparticles or nanorods [6–9], and other custom nanoparticles [10]. These contrast agents are highly scattering and typically provide increased reflectivity. Conversely, methylene blue has been used as an absorptive contrast agent [11,12], and spectral contrast has been investigated using indocyanine green (ICG) ex vivo [13] and gold nanorods in vivo [9].

Here we present how ICG, a common fluorescent contrast agent approved by the United States Food and Drug Administration for intravenous injection in humans, can provide absorption and spectral contrast for OCT imaging in the mouse eye in vivo. ICG is routinely used in ophthalmology for fluorescence angiography [14]. When bound to blood plasma, ICG absorbs light strongly around 800 nm (Fig. 1) and emits fluorescence at longer wavelengths [15]. Because the emitted fluorescence is not coherent, it does not contribute to the modulated OCT signal and is not used in this application. Rather, the absorptive and spectral properties of the tracer are used. After comparing the efficacy of absorption and spectral contrast mechanisms, we evaluate the stability of spectral contrast as compared to traditional intensity-based measurements.

In this study, the eyes of three very-low-density-lipoprotein receptor knockout mice (${\rm VLDLR}^{-/-}$, age 11 weeks), an established model of retinal neovascularization [16], and three wild type controls (age 96 weeks) were imaged with OCT before, during, and after a 3 mL/kg intravenous bolus injection of 5 mg/mL ICG (VERDYE, Diagnostic Green GmbH), delivered via the tail vein. This is the same concentration of ICG used clinically, but with a larger relative volume. Animals were anesthetized with isoflurane, using a 4% dose for 4 min to induce anesthesia and a 2% dose throughout the imaging procedure to maintain the depth of anesthesia. Pupils were dilated with Tropicamide (5 mg/mL, AGEPHA Pharma s.r.o.), and artificial tears (Oculotect, ALCON Pharma GmbH) were used to prevent drying of the eyes. All animal protocols have been approved by the Austrian Ministry of Education, Science, and Research (BMBWF/66.009/0272-V/3b/2019).

Imaging of the retinas was performed using a custom OCT ophthalmoscope for rodent imaging [17] with a superluminescent diode source (Superlum D-840-HP) centered at 840 nm with a 100 nm bandwidth, yielding an axial resolution of 5.1 µm in air ($\sim{3.8}\;{\unicode{x00B5}{\rm m}}$ in tissue). The source spectrum (Fig. 1) was measured with an optical spectrum analyzer (AQ-6315E, ANDO). Dynamic contrast OCT (DyC-OCT) time courses [3] were obtained by repeatedly scanning the same two-dimensional (2D) cross-section over time for 15 s at a 130 Hz B-scan rate during the injection of the ICG contrast agent. Volumetric scans with five B-scan repeats were also acquired before and after injection. The three-dimensional (3D) images covered a ${1}\;{\rm mm}\; \times \;{1}\;{\rm mm}$ region centered on the optic nerve head.

 

Fig. 1. Spectral dependence of indocyanine green (ICG) as an OCT contrast agent. The approximate absorption spectrum of ICG under predicted in vivo conditions (green) [15] is shown alongside the illumination spectrum of the OCT light source (black). This spectrum is Gaussian filtered in post-processing (dotted lines) to highlight short (blue), long (orange), or all wavelengths (black).

Download Full Size | PDF

 

Fig. 2. DyC-OCT signals following tracer injection. (A) Intensity time courses at the same choroidal region (Fig. 3, green box) following the ICG injection using the spectral filters in Fig. 1. Time courses were normalized by the baseline signal before arrival of the tracer. (B) Relative change in signal following the ICG injection for the inverted intensity (black) and spectral ratio (purple), defined as the ratio between the intensities measured with the long (orange) and short (blue) spectral windows. The total intensity was inverted in B to provide a fair comparison with the SR signal.

Download Full Size | PDF

Both traditional OCT intensity images and spectrally sensitive OCT images were generated from the data. Traditional OCT processing used a broad Gaussian window centered on the source spectrum to reduce side lobes in the point-spread function (Fig. 1). Spectral processing used two narrower Gaussian windows, similar to other spectrally sensitive OCT methods [9,18], each half the width of the traditional window in k-space, located at the peak of and away from the ICG absorption to generate intensity images that were either highly sensitive or insensitive to the contrast agent (Fig. 1). Both the traditional OCT intensity (I_t) images and the spectrally sensitive images were normalized in the linear scale by dividing by the mean noise intensity in the vitreous. A new metric called the spectral ratio (SR, Fig. 2) was measured for each pixel in the retinal cross section (Fig. 3) as the longer-wavelength, ICG-insensitive intensity (${{I}_i}$) divided by the shorter-wavelength, ICG-sensitive intensity (${{I}_s}$). It is important to note that when averaging or projecting the spectral data, we averaged ${{I}_t}$, ${{I}_i}$, and ${{I}_s}$ separately before performing further operations to reduce noise contributions in the end result, i.e., $\overline {\rm SR} = {\bar I_i}/{\bar I_s}$. All datasets were motion corrected and flattened identically, and spectral datasets were axially aligned using cross-correlation methods prior to SR calculations.

 

Fig. 3. Absorption contrast visible from traditional OCT intensity processing (left column) is compared with the spectral contrast provided by the spectral ratio (right column) in the retina of a control mouse. The first row shows the respective signals before the ICG injection, and the second row shows the signals at the peak of the first passage. Each of these images is temporally averaged 200 times over $\sim{1.5}\;{\rm s}$. The last row shows the relative change in intensity and SR signals between the two time points in the first two rows. The relative change in intensity is given as the intensity before injection divided by the intensity after injection, due to the drop in intensity, to provide a fair comparison with the relative change in spectral ratio which increases following injection. The green boxes show the location of the signal time courses plotted in Fig. 2. IR, inner retina; RPE, retinal pigment epithelium; Ch, choroid.

Download Full Size | PDF

The injection of ICG induces a decrease in the OCT intensity in the highly vascular choroid due to increased absorption (Fig. 2A). This decrease becomes more pronounced when the short spectral filter is used and is largely invisible when the long spectral filter is used. The SR signal was computed from the two spectral channels and compared against the total intensity, which is inverted here to provide a fair comparison (Fig. 2B). SR was more sensitive to the ICG injection, revealing secondary and tertiary peaks caused by recirculation of the tracer. To quantitatively evaluate the signal quality in the DyC-OCT time courses, the signal-to-noise ratio (SNR), defined as the mean signal at the peak of the first bolus passage divided by the standard deviation of the baseline signal, and the contrast-to-noise ratio (CNR), defined the same as SNR, but using the baseline-subtracted signal, were used. In Fig. 2, inverted total intensity had an SNR of 7.0 and a CNR of 1.0, whereas SR had an SNR of 39.2 and a CNR of 15.1. Similar improvements were observed throughout the choroid.

Intensity and SR contrasts were compared across the same 2D cross-section, both before and after the ICG injection (Fig. 3). Changes in intensity following the ICG injection are minimal, particularly when using the traditional log-scale visualization to accommodate the large dynamic range. The SR, by comparison, has a much more manageable dynamic range and demonstrates larger changes following injection, particularly in the choroid; however, this comes at the cost of half the axial resolution, due to the narrower Gaussian windows used for spectral image processing. The relative change map gives a direct comparison of the two signal types in linear scale, demonstrating higher sensitivity to the tracer using the SR signal, particularly in the choroid. The large relative change around the retinal pigment epithelium (RPE) is likely caused by slight motion between the baseline and peak time points rather than by ICG contrast.

Next, we investigated the SR signal and its response to ICG injection in full 3D volumes (Fig. 4). The inner retina, RPE, and choroid were each segmented using automated methods, and en face projections were generated. A clear increase in the SR signal within the large vessels of the inner retina is observed. While some shadowing occurs in the RPE following injection, the appearance of the RPE itself remains consistent between the scans. Surprisingly, similar shadowing is not apparent in the inner retina directly below the large superficial vessels (top row, cyan). A widespread increase in SR is observed in the highly vascular choroid, revealing large, deep vessels (bottom row, cyan).

 

Fig. 4. Mean intensity projections of the spectral ratio across the inner retina, retinal pigment epithelium (RPE), and choroid of a VLDLR mouse before and after the injection of ICG, with a time separation of 7 min. Brightness denotes the measured SR and colors roughly code depth, where red is superficial, and cyan is deep for each region (see colorbar labels). White arrows mark lesion sites in the superficial RPE and large vessels in the deep choroid. Green circles mark the lateral bounds of the cylindrical shell used for averaging in Fig. 5.

Download Full Size | PDF

In addition to the vascular SR signal contributions, which increased following injection, there were also substantial SR signals that did not change following injection. The strong SR signal and granular pattern of the RPE may be related to melanin distribution [19]. The strong SR signals in the outer nuclear layer above the RPE (middle row, red) are likely caused by the migration of melanin particles around lesion sites, previously demonstrated in this animal model with polarization-sensitive OCT [20]. These deposits were visible around lesions in all VLDLR animals, but none of the control animals. A cylindrical shell around the optic nerve head, with an inner radius of 150 µm and an outer radius of 500 µm, was further averaged to provide depth profiles of intensity and SR (Fig. 5).

 

Fig. 5. Traditional intensity and SR processing was applied to 3D volumes, and laterally averaged depth profiles were measured within a cylindrical shell (Fig. 4, green). (A) Log-scale intensity profiles and (B) linear-scale relative intensity change profiles reveal a global decrease over time. (C) SR profiles and (D) relative SR change profiles reveal an increase in the SR signal only in highly vascular regions (*) following the ICG injection.

Download Full Size | PDF

Intensity profiles demonstrated a clear decrease over time across all regions of the retina (Figs. 5A and 5B). The general drop in intensity between volumes is not likely caused by the ICG tracer itself, but rather by evaporation of the tear film, realignment of the beam, or other factors that can reduce the amount of light delivered to or collected from the retina. Despite these large intensity changes, the SR remains stable as a function of depth, with some increases noted following the injection of ICG (Figs. 5C and 5D). We attribute these increases to the circulation of the ICG tracer as they correspond to the locations of large vessels in the inner retina and to the highly vascular choroid. These SR changes also follow the expected tracer decay pattern, with an increase due to the injection and a subsequent decrease, due to removal of the tracer from the circulating blood. Note that the SR in the RPE and the outer nuclear region directly above the RPE, which are avascular and thus should be unaffected by the ICG injection, do not substantially change following tracer administration. SR changes due to the tracer injection are easily distinguishable from the background noise and are clearly depth resolved (Fig. 5D). Due to the solvent- and concentration-dependent absorption profile of ICG [15], further studies are required to determine the most efficient and lowest viable doses for SR imaging.

It makes sense to compare our SR metric with the MOZART metric (for molecular imaging and characterization of tissue noninvasively at cellular resolution), as they are both spectrally sensitive and can be computed from the same data. The MOZART metric, previously applied to gold nanorods [9], is defined as the difference of the spectral intensities divided by the total intensity. Similar to SR, we have computed the mean MOZART signal as $\overline {\rm MOZART} = ( {\overline {{I_i}} - \overline {{I_s}} } )/\overline {{I_t}} $. DyC-OCT time courses for SR and MOZART were highly similar in shape; however, MOZART proved to be noisier, yielding an SNR of 25.6 and a CNR of 9.0 for the same region plotted in Fig. 2. While these values are a substantial improvement over the intensity signal, SR provided a 65% higher SNR and an 87% higher CNR than MOZART when averaged across the choroid.

An axial profile was also computed for MOZART (Fig. 6). Similar to SR, MOZART remained fairly stable over time and increased in the vascular regions of the retina following ICG injection. The relative changes from baseline, however, were noisier, making it more difficult to distinguish static tissue from vascular regions. In particular, the choroid produced a much less pronounced change in the MOZART signal compared to SR.

 

Fig. 6. MOZART processing was applied to 3D volumes, and laterally averaged depth profiles were measured within a cylindrical shell (Fig. 4, green). (A) Average MOZART profiles and (B) relative MOZART change profiles before and after tracer injection. Increases are observed in highly vascular regions (*) after injection.

Download Full Size | PDF

While the spectral ratio between long and short channels (Fig. 2) does not further amplify the spectral contrast beyond that offered by the short channel, the longer-wavelength ICG-insensitive channel does provide a valuable temporally and spatially co-registered reference that can correct depth-wise changes in intensity over time, for example due to changes in focus or cataract formation, leading to highly stable SR measurements. Quantitatively co-registering data is easier for DyC-OCT imaging [4] when the imaging time is short, and the sample remains stable, but it becomes a significant problem when comparing images acquired days, hours, or even minutes apart. The SR metric is highly consistent across volumes acquired up to 9 min apart (Figs. 4, 5C and 5D), requiring no further normalization despite substantial changes in intensity (Figs. 5A and 5B) and complete refocusing and alignment of the mouse eye between volumes.

SR imaging of ICG contrast-enhanced OCT may be able to provide similar information to ICG fluorescence angiography, but with the added benefit of depth resolution and perfectly co-registered OCT intensity and angiography data. Traditional FA is not able to directly distinguish contributions from different vascular layers and relies on time-course information [14] to separate the choroid from the inner retina. Using the depth-resolved SR, however, the choroidal contributions can clearly be separated from the inner retina (Figs. 5C and 5D). While other groups are evaluating hardware-based solutions to provide depth resolution to FA [21], the method presented here is simple and may be integrated in post-processing with current clinical OCT systems and clinically relevant tracers such as ICG.

Here we demonstrated absorptive and spectral contrast using ICG as an OCT contrast agent in the mouse eye in vivo. The ratio of ICG-sensitive and ICG-insensitive OCT intensities, acquired in post-processing from a single OCT volume, provides both high sensitivity to ICG and high stability across time points set up to 9 min apart despite large differences in intensity. This new method may assist in imaging and diagnosing changes in the choroid, which has historically been difficult to study using traditional OCT methods.