The inner limiting membrane (ILM) is a thin, transparent membrane that forms a boundary between the vitreous and retinal nerve fiber layer. Surgical peeling of the ILM benefits post-operative outcomes in patients with macular hole, epiretinal membrane, and some forms of chronic diabetic macular edema [1]. Exogenous dyes, such as indocyanine green (ICG), are routinely used to preferentially stain the transparent ILM and enhance contrast under white light microscopy [2–4]. Optical coherence tomography (OCT) for intraoperative imaging during ophthalmic surgery has enabled real-time, depth-resolved visualization of retinal structures and instrument-tissue interactions without the need for exogenous contrast [5–8]. However, the ILM is difficult to distinguish from the underlying retina on OCT images, limiting the utility of OCT to guide ILM peeling procedures.

While ICG is used because it preferentially stains the ILM [2], OCT does not have a contrast mechanism to detect ICG and identify the ILM. Any increase in OCT signal intensity due to ICG is nonspecific and cannot be reliably detected. Furthermore, studies have observed retinal phototoxicity at clinical concentrations of ICG (0.6–5 mg/cc) [9,10], further highlighting the need for high-sensitivity detection of ICG during ophthalmic surgery.

Molecular OCT [11] is a functional extension of OCT that adds molecular contrast, and includes modalities such as spectroscopic [12], pump-probe [13], magnetomotive [14] and photothermal OCT (PT-OCT) [15,16]. These techniques could provide specific detection of the ICG-stained ILM within the OCT field-of-view. However, to our knowledge no molecular OCT technique has been used to detect ICG in the eye.

We propose photothermal PT-OCT to specifically detect the ICG-stained ILM. PT-OCT detects optical absorbers in the OCT field-of-view by coupling an additional amplitude-modulated laser to a phase-sensitive OCT system. Light from the photothermal laser is absorbed in the sample, producing a small local increase in temperature that results in a change in refractive index and elastic expansion of the tissue. Both effects change the optical path length, detected as a change in the OCT phase. The location of absorbers is determined in post-processing based on the phase oscillations at the modulation frequency of the photothermal laser. The PT-OCT signal is quantitative and proportional to variables such as absorber concentration and photothermal laser power, as previously demonstrated [17],

In the past, PT-OCT has mostly been used to detect gold nanoparticles in phantoms [15], cells [18], and bulk tissue [19]. PT-OCT has also been used to detect high concentrations of ICG ($>0.5\mathrm{mg}/\mathrm{cc}$) ex vivo in phantoms [20] and in vivo in the mouse ear [21] using high photothermal laser powers (70–80 mW). Additionally, PT-OCT has been demonstrated in vivo in the mouse eye to detect melanin and gold nanorods [22]. In this Letter, we demonstrate PT-OCT of ICG in phantoms at low ICG concentrations (0.08–0.6 mg/cc) and using low photothermal laser powers (0.4–3 mW, ANSI maximum exposure $\sim 0.5\mathrm{mW}$ [23]). We also demonstrate PT-OCT of ICG in cadaveric pig eyes to identify the ILM and perform an ILM peel.

A commercial spectral domain OCT (Bioptigen, $\lambda =860\mathrm{nm}$, 93 nm bandwidth, 36 kHz acquisition rate) was modified to perform PT-OCT. A diode laser (Coherent, $\lambda =685\mathrm{nm}$, $f_0=500\mathrm{Hz}$ amplitude-modulation, square wave, 50% duty cycle) was added to the light path via a 50:50 coupler to provide photothermal excitation. A circulator (AC Photonics) was also added to redirect the OCT signal from the sample to the spectrometer. Excitation at 685 nm corresponds to the absorption peak of ICG at high concentrations in water [24]. The OCT superluminescent diode originally had a power of 0.7 mW at the sample, but the addition of the circulator reduced the power to 0.470 mW at the sample. The photothermal laser had a power of 0–3.12 mW at the sample depending on the experiment. B-scans were acquired across 2.5 mm with 400 pixels, each consisting of 700 repeated A-scans (M-scan). Each B-scan took approximately 7 s to acquire. PT-OCT data processing was performed after data acquisition as previously described using a custom MATLAB code [17]. In summary, the data was first resampled from wavelength to wavenumber, dispersion corrected [25], and background subtracted. A Chirp-$Z$ transform was used to convert the wavenumber dimension into a spatial domain and obtain the OCT intensity and phase signal. For each pixel in depth over the 700 M-scans, the first temporal derivative of the phase data was calculated. A Fourier transform was used to convert the phase oscillations from the time domain to the frequency domain and allowed identification of the peak corresponding to the photothermal modulation frequency ($f_0=500\mathrm{Hz}$). The amplitude of the peak was taken to be the PT-OCT signal intensity as seen in Eq. (2) [17],

A constant 0.5 nm noise at a frequency between 503 and 505 Hz was present in all negative controls, phantoms, and tissue. This created a beat frequency in all positive controls, both in phantoms and tissue. Thus, a 0.5 nm constant signal at 503–505 Hz was subtracted from all images for all samples to improve visualization.

Phantoms were made from 1 μL of ICG diluted in saline at concentrations 0.08, 0.16, 0.20, 0.31, 0.39, and 0.62 mg/cc placed on colored label tape and imaged immediately. Samples were imaged with the photothermal laser power set at 0 (laser off), 0.48, 0.59, 0.80, 1.0, 1.2, 1.7, 2.3, and 3.4 mW measured at the sample.

Freshly excised cadaveric pig eyes were obtained for imaging ($\sim 6\mathrm{h}$ after death). To help separate the anterior hyaloid membrane from the lens, air (0.5 cc) was injected into the vitreous cavity through the sclera via a 30 G needle 2 mm posterior to the temporal limbus. The air bubble was left for 10–30 min. A 15° blade was used to make an incision at the corneoscleral limbus and then enlarged by curved Stevens scissors to completely remove the cornea. A lens loop and muscle hook were used to remove the crystalline lens. A syringe with an 18 G flat-end fill needle was used to engage vitreous with aspiration, lift in anteriorly, and then severe its attachments with Wescott scissors. This engaged and separated the posterior hyaloid membrane from the retinal surface. The retinal surface was then dried with a Weck-Cel sponge applied to the optic nerve head.

ICG was diluted at concentrations of 0.25 and 0.5 mg/cc in saline and kept wrapped in foil until application ($<15\min$) to reduce photobleaching. The ICG was applied to the ILM for 1 min. The retina was then rinsed with saline and all liquid above the ILM was removed using Weck-Cel sponges applied to the optic nerve head. The retina was then imaged with PT-OCT.

To validate that the ICG was specifically staining the ILM, an ILM-peel was performed on one ex vivo pig eye. The eye was prepared as previously described and stained with 0.25 mg/cc ICG for 1 min before being rinsed with saline and dried with a Weck-Cel sponge. A section of the ILM was then peeled using 23 G Eckardt-style ILM forceps under a white light surgical microscope.

The average PT-OCT signal obtained in phantoms as a function of photothermal laser power can be seen in Fig. 1(a) for different ICG concentrations and a saline control (0 mg/cc). In accordance to theory [Eq. (1)], the PT-OCT signal increases linearly with the photothermal laser power ($R^2>0.98$) for samples with low ICG concentration ($<0.3\mathrm{mg}/\mathrm{cc}$). Samples with higher concentrations (0.31–0.62 mg/cc) cannot be considered optically transparent ($P(z)\ne P_0$) and thus do not follow Eq. (1). The nonlinear increase in PT-OCT signal is likely caused by shadowing (due to the high absorption coefficient) [27] and reabsorption of emitted photons [28]. The absorption coefficient of ICG is not always linearly proportional to ICG concentration [24], but the PT-OCT signal does generally increase with ICG concentration.

 

Fig. 1. (a) Average PT-OCT signal as a function of photothermal laser power for all ICG phantom concentrations. Dashed lines: linear fit for low concentration samples. Error bar: standard deviation over one B-scan. (b) PT-OCT signal for 0.16 mg/cc of ICG (orange) in comparison to saline (blue) as a function of laser power. (c) PT-OCT signal as a function of ICG concentration using 0.48 mW (orange) or 0 mW (blue) of laser power. Linear fit (solid lines), with 95% confidence interval (dotted lines). ${}^{*}p<0.05$, ${}^{****}p<0.001$ difference between slopes.

Download Full Size | PDF

To determine the lowest detectable ICG concentration, the slope of each linear fit for the ICG samples in Fig. 1(a) was compared to the slope of the saline sample. As seen in Fig. 1(b), the slope of the 0.16 mg/cc ICG sample ($\text{slope}=0.203\pm 0.006$) was significantly different from the slope of the saline sample ($0.012\pm 0.004$) ($p<0.001$, ANCOVA [29]). Similarly, to determine the lowest usable photothermal laser power, the average PT-OCT signal as a function of ICG concentration was plotted for each laser power, and a linear fit was performed. As seen in Fig. 1(c), the slope of the distribution acquired at 0.48 mW ($\text{slope}=0.24\pm 0.06$) is significantly different from the slope of the distribution acquired with the laser off ($0.02\pm 0.03$) ($p<0.05$, ANCOVA [29]). The 0.08 mg/cc sample was not different from control when imaged at 0.48 mW.

The PT-OCT signal of the saline sample [Fig. 1(b), blue] increases by a very small amount as a function of photothermal laser power ($\text{slope}=0.012\pm 0.004$), possibly due to light absorption by the saline or tape. The PT-OCT signal does not increase as a function of ICG concentration [Fig. 1(c), blue] when the laser is turned off ($\text{slope}=0.02\pm 0.03$). No photobleaching of ICG was detected over one A-scan, and B-scans were taken at different sample locations to minimize photobleaching. Based on these results, we conclude that PT-OCT is sensitive to ICG concentrations as low as 0.16 mg/cc while using photothermal laser powers as low as 0.48 mW in phantoms.

PT-OCT B-scans were acquired in ex vivo pig eyes to better demonstrate our technique. OCT [Fig. 2(a)] and corresponding PT-OCT B-scans [Figs. 2(b) and 2(c)] of the retina are shown after application of 0.5 mg/cc of ICG. The photothermal laser was set to 0.89 mW for Fig. 2(b), and the PT-OCT signal can be seen at a location consistent with the ILM [Fig. 2(d)]. The same location on the retina was also imaged with PT-OCT at a higher laser power [3.12 mW, Figs. 2(c) and 2(e)] to confirm the presence of the PT-OCT signal all along the ILM. For comparison, no PT-OCT signal was detected when the PT-OCT B-scan of a retina at 3.12 mW without ICG application was acquired (data not shown). This confirms that PT-OCT can detect the ICG-stained ILM at low laser powers.

 

Fig. 2. PT-OCT of an ex vivo porcine retina after the ILM was stained with 0.5 mg/cc of ICG. (a) OCT (gray) and (b) corresponding PT-OCT B-scans (overlaid in green) using 0.89 mW or (c) 3.12 mW of photothermal laser power. Magnified view of the boxed area in (d) and (e). Scale bar: 200 μm.

Download Full Size | PDF

An ILM peel was performed in one eye after 0.25 mg/cc ICG application to demonstrate that the ICG dyes only the ILM. In the clinic, this procedure would normally be performed under a white light surgical microscope, where ICG is used to increase the contrast between the ILM and the retina. An image acquired using the microscope camera after ICG application and ILM peel can be seen in Fig. 3(a). The boundary between the area where the ILM was peeled and the area where the ILM is intact is indicated by the white arrows. The ICG concentration used is low compared to the concentrations traditionally used in surgery for this procedure (0.5–5 mg/cc), explaining the very low contrast between ILM and non-ILM regions in this image when visualized under white light. PT-OCT B-scans perpendicular to the boundary were acquired, and an example OCT (white-dashed line) and corresponding PT-OCT B- scans can be seen in Figs. 3(b) and 3(c). The PT-OCT B-scan was acquired with a photothermal laser power of 3.12 mW. A clear PT-OCT signal can be seen where the ILM was left intact [Fig. 3(d)], while no PT-OCT signal above noise floor can be seen in the region where the ILM was peeled. Presence and absence of ILM was determined based on the surgeon’s feedback and the morphology of the retina as seen on OCT [Fig. 3(b)].

 

Fig. 3. PT-OCT of an ILM-peel in ex vivo pig eye. (a) White light image from the surgical microscope showing the boundary between the intact ILM dyed with 0.25 mg/cc ICG and the peeled ILM (arrows) with the B-scan location (dashed line). (b) OCT (gray) and (c) corresponding PT-OCT B-scan (green) showing a cross-section of the retina with intact ILM and after it was peeled (arrows at boundary). (d) Magnified view of boxed area in (c). Scale bar: 200 μm.

Download Full Size | PDF

OCT is being incorporated into the surgical suite to provide depth-resolved visualization of instrument-tissue interaction during retinal surgery [8]. Peeling the ILM is one such procedure where OCT provides crucial information to the surgeon. To increase the contrast between the ILM and other layers of the retina, surgeons routinely use dyes such as ICG [1], which specifically binds to the ILM [2]. However, OCT lacks a contrast mechanism to detect ICG, and the high concentrations of ICG that are used in surgery may be associated with phototoxicity.

In this study, we demonstrated that PT-OCT can detect low concentrations of ICG used to stain the ILM in ex vivo pig eyes. This is the first time a molecular OCT technique has been used to specifically detect ICG in the eye.

In the first part of this study, we tested the sensitivity of PT-OCT to low concentrations of ICG at low photothermal laser powers. In the past, PT-OCT was performed at high photothermal laser powers (70–80 mW) [20,21] to increase the signal intensity when using ICG as a contrast agent. The ANSI maximum permissible exposure for a laser at 685 nm for a 7 s scan is approximately 0.5 mW [23]. In this Letter, we demonstrate in phantoms that PT-OCT is sensitive to ICG at a clinically relevant laser power (0.48 mW) and low concentrations of the dye (0.16 mg/cc) that should reduce potential phototoxicity.

In the second part of this study, we detected the presence of ICG on the ILM of cadaveric pig eyes using PT-OCT. For this experiment, the cornea, lens, and vitreous were removed to expose the ILM. In human patients, the vitreous can be removed and replaced with saline while keeping the cornea and lens in place and without damaging the ILM. The ICG can then be injected into the eye to stain the ILM. However, in pigs the vitreous is strongly attached to the ILM, preventing both a normal vitrectomy or an ICG injection in a whole cadaveric eye [30]. For that reason, we gently removed the vitreous with Weck-Cel sponges to avoid any damage to the ILM or the retina before ICG application and imaging. In the future, this experiment could be repeated in an in vivo pig model, where a traditional vitrectomy and ICG injection could be performed before PT-OCT imaging. This would more closely resemble the clinical surgical procedure.

Finally, we performed an ILM peel in a cadaveric pig eye to confirm our hypothesis that ICG specifically stained the ILM and that peeling the ILM would remove all of the PT-OCT signal due to the ICG. In the images obtained during this experiment, a clear PT-OCT signal can be seen where the ILM is intact, and no PT-OCT signal above the noise floor is present where the ILM has been peeled, confirming our hypothesis. This also clearly demonstrates how ICG combined with PT-OCT imaging would improve contrast of the ILM for intraoperative OCT guidance.

In the future, strategies to reduce PT-OCT imaging times will be evaluated to increase the B-scan rate and acquire volume scans. Reducing the number of repeated A-scans per pixel would increase imaging speed, especially if combined with a faster photothermal modulation frequency. For example, 200 repeated A-scans instead of 700 would reduce the acquisition time to 2 s, which is the same acquisition time as a Doppler scan on our Bioptigen system.

In conclusion, we detected clinically low concentrations of ICG ($<0.25\mathrm{mg}/\mathrm{cc}$) using PT-OCT at low photothermal laser powers ($<0.5\mathrm{mW}$). Additionally, we performed an ILM peel in ex vivo pig eyes to demonstrate that the PT-OCT signal is specific to the ICG-stained ILM. These preliminary results show the potential of PT-OCT for clinical use in the eye.