1.

Introduction

Selective retina therapy (SRT) is an effective laser treatment method for various retinal diseases associated with degradation of the retinal pigment epithelium (RPE), such as diabetic macular edema, central serous chorioretinopathy, and age-related macular degeneration.^1–6 The RPE, which contains a high concentration of melanosomes, absorbs 50% to 60% of incident green light. However, in order to selectively target the RPE layer, the laser pulse duration needs to be shorter than a thermal relaxation time of the RPE ($\sim 10\mu \mathrm{s}$).⁷ The SRT reduces negative side effects and facilitates healing of the induced retinal lesions by avoiding thermal damages of the adjacent photoreceptors, the neural retina, and the choroid. However, the selection of proper laser energy—which is crucial for successful SRT without excessive burning and collateral damage—is challenging because lesions in the RPE are ophthalmoscopically invisible. In addition, different melanin concentrations among patients, or regions even within an eye,⁸ make it impossible to set a static threshold value of pulse energy of a therapeutic irradiation window.

Fundus fluorescence angiography (FFA) is an accurate method to detect the lesions, but it requires the use of fluorescent dye injection⁹ and a long delay between treatment and detection. For real-time noninvasive SRT monitoring, several approaches have been proposed. These include the detection of microbubble formation and collapse, which induce mechanical disruption and damage to RPE cells.¹⁰ This approach measures the acoustic transient¹¹ or light reflection changes.¹² Although these methods have already been used in several clinical studies,^11,13,14 they do not provide visual feedback during the treatment. Optical coherence tomography (OCT), which can provide depth-resolved imaging, was also applied for the SRT monitoring.^15–19 The treatments are considered successful when OCT signal variations, i.e., intensity decrease, are detected, and the results show good agreement with the evaluation of lesions by FFA.

Speckle variance OCT (svOCT) quantifies the speckle pattern variation caused by moving particles or structural changes in biological tissues. It calculates the interframe intensity variance of a sequence of structural OCT images. The svOCT has been extensively developed in recent years for OCT angiography, which is used to visualize retinal microvasculatures;²⁰ it has also been shown to be able to monitor protein denaturation and coagulation.²¹ Thus, it is expected that svOCT could be an effective way to detect speckle variation changes induced by morphological and structural changes of retinal tissue during the thermal-induced microbubble formation and collapse by laser irradiation. In this paper, we studied and demonstrated SRT monitoring based on the svOCT. A swept-source OCT imaging system integrated with a microsecond pulsed laser system was used for ex vivo bovine retina study. SvOCT images corresponding to various laser pulse energies and the various number of frames were obtained. The svOCT values of RPE and photoreceptor layers were averaged along the axial direction, and peak values of the svOCT at each pulse laser irradiation were analyzed. The microscopic images of the treated spots were taken before and after removing the upper neural retinal layers to assess the degree of retina and RPE damage. Spatial and temporal thermal effects in retina induced by pulse laser irradiation were simulated and correlated to the peak values of svOCT.

2.

Experimental Method

An in-house built swept-source OCT imaging system was integrated with a frequency-doubled Nd:YLF laser-based SRT system (Lutronic, Goyang, Korea). The schematic of the system is shown in Fig. 1. The OCT system used a commercial swept-source engine (Axsun Technologies Inc., Billerica, Massachusetts) operating at 100-kHz sweep rate. The center wavelength and sweeping bandwidth of the system were 1060 and 100 nm, respectively. The OCT laser was combined with the pulse laser using a dichroic mirror; galvano mirrors were used to direct the OCT laser to a treated spot on the retina. The wavelength of the pulse laser was 527 nm, and the pulse laser operated at 100-Hz repetition rate and $1.7\text{-}\mu \mathrm{s}$ duration. The pulse laser energy was adjusted from 22 to $190\mu \mathrm{J}$ using neutral-density filters. In this study, fresh ex vivo bovine eyes were acquired from a local butcher and immersed in a cooled saline solution. Bovine eyes have tapetum fibrosum, which has the retinal epithelial layer completely unpigmented, over the central and mid-region of the retina.²² Because the RPE layer exists in the periphery of the retina, it was difficult to focus the beam on the RPE layer using the crystalline lens of an eye itself. Therefore, the bovine eye’s cornea and lens were removed, and the beam was focused on a tilted eye using an objective lens. A total of 39 treated spots were tested on two eyes. M-mode OCT images of the bovine retina were acquired during the laser pulse irradiation.

Fig. 1

Schematic of a swept-source OCT system integrated into a pulse laser system. BD, balanced detector; OL, objective lens.

SvOCT images were calculated by

Spatial and temporal temperature distributions in the retina were numerically calculated by COMSOL Multiphysics software. Spatial distribution was calculated in 2-D, and the geometry of the bovine retina model used for the simulation is shown in Fig. 2(a). The RPE was modeled as a $7\text{-}\mu \mathrm{m}$ layer containing melanosomes that were assumed as spheres of radius $0.3\mu \mathrm{m}$ by a discrete absorber model.²³ Melanosomes were diagonally distributed, and the distance between adjacent melanosomes was set to $1.2\mu \mathrm{m}$. The model applied a heat equation as shown below:

Fig. 2

(a) The geometry of the bovine retina model. The retina was assumed to consist of two layers, neural retina and RPE, and has immediate contact with a choroid. The RPE was modeled as a $7\text{-}\mu \mathrm{m}$ layer containing melanosomes that were assumed as diagonally distributed spheres of radius $0.3\mu \mathrm{m}$. (b) Temperature–time dependence in the neural retina and at the melanosome surface in RPE when laser pulse of energy $50\mu \mathrm{J}$ irradiated. Spatial distribution of temperature around RPE, when the temperature of the neural retina reached a maximum after (c) $20\text{-}\mu \mathrm{J}$, (d) $50\text{-}\mu \mathrm{J}$ and (e) $100\text{-}\mu \mathrm{J}$ pulse irradiations.

Table 1

Thickness, absorption coefficient, and thermal properties of each retinal layer.

The temperature of the neural retina was calculated by averaging the temperature in a rectangular region ($30\mu \mathrm{m}\times 20\mu \mathrm{m}$) $2.5\mu \mathrm{m}$ away from RPE, and the temperature of the melanosome surface in RPE was calculated by averaging the temperature on seven melanosome surfaces at the first two melanosome layers located in the center of the Gaussian beam. Figure 2(b) shows the time-dependent temperature variation in the neural retina (in orange) and at the melanosome surface in RPE (in blue). Spatial distributions of temperature around RPE, when the temperature of neural retina reached a maximum after a 20-, 50-, and $100\text{-}\mu \mathrm{J}$ pulse irradiation, are shown in Figs. 2(c)–2(e). The calculated peak temperatures in each region were correlated to the peak values of svOCT and tissue damage.

3.

Results

Figures 3(a) and 3(b) show an M-scan OCT image and the corresponding svOCT image of the bovine retina when $108\text{-}\mu \mathrm{J}$ energy per pulse irradiated. The M-scan OCT image shows the visible temporal signal variations induced by the laser pulse at the moment marked with white triangles. The signal variation increases the svOCT value. Figure 3(c) shows svOCT values averaged in ROI, and it shows a distinctive peak for each pulse irradiation.

Fig. 3

(a) M-mode OCT image of the bovine retina and corresponding (b) svOCT image. Photoreceptor and RPE layers, which are highly scattering and absorptive, were set as an ROI. (c) Axially averaged svOCT values in the ROI during pulse laser irradiation. White triangles mark the moment when each laser pulse ($108\mu \mathrm{J}$) irradiated.

To find appropriate $N$ for calculating the svOCT image, we first tested different values of $N$ and compared the average peak values of the svOCT spikes, the standard deviation of those peaks, and background noise level. Since the speckle variation induced by each laser pulse lasts for only around $50\mu \mathrm{s}$, only five frames of the speckle variances show high signal while the rest show very low signal. Intuitively, if we choose a larger window size $N$, the lower portion of sv signal in the window will be significant, and it will decrease the overall value of the sv signal inside the window, i.e., the peak value of svOCT spikes. Furthermore, the variance of those peak values will also decrease. Figures 4(a)–4(d) show how the average peak values (shapes) and standard deviation (error bar) change depending on laser energy level when $N$ is 2, 5, 10, and 20, respectively. We can see that, as expected, both the average peak value and the standard deviation decrease when $N$ increases. Relative standard deviation, defined as the ratio of the standard deviation to the mean, was 0.53, 0.46, 0.43, and 0.46 for $N$ of 2, 5, 10, and 20, respectively. Therefore, a mid-range of $N$ values between 5 and 10 are suitable for calculating svOCT in terms of the precision and repeatability. In addition, note that the background noise level increases with increasing $N$. For $N$ of 2, 5, 10, and 20, the average upper bound levels of background noise were $0.54\pm 0.24$, $0.74\pm 0.31$, $1.03\pm 0.41$, and $1.64\pm 0.66$, respectively. The background noise levels were bounded by $\mu +3\sigma$ upper limit, where $\mu$ was svOCT values averaged in ROI before or after laser irradiation and $\sigma$ was the standard deviation. The increase in background noise was caused by the bulk motion of the sample and other environmental changes. Shorter integration time will decrease these effects and so will the smaller value of $N$. Considering both the effects of relative standard deviation and background noise levels, we choose $N$ as 5.

Fig. 4

Mean (shapes) and standard deviation (error bar) of peak values of svOCT values averaged in ROI depending on laser pulse energy when window size $N$ is (a) 2, (b) 5, (c) 10, and (d) 20.

Figure 5 shows svOCT values averaged in ROI depending on pulse laser energy when $N$ is 5. Peak values increased with increasing pulse laser energy, and the distinctive peaks were observed when the laser pulse energy was as low as $54\mu \mathrm{J}$. Figures 6(a) and 6(b) show microscopic images of the treated spots before and after peeling upper neural retinal layers off, respectively. The energy level of each spot is shown in Fig. 6(c). The denaturation of neural retina can be observed as whitish spots (pointed by white triangles) in Fig. 6(a), but the lesions confined only to the RPE layer are invisible. The leftmost and rightmost columns are high-energy lesions marking pattern of the spots. In Fig. 6(b), lesions in the RPE layer (pointed by arrows) can be detected by peeling off the upper neural retinal layers.

Fig. 5

SvOCT values averaged in ROI when pulse laser energy is (a) 150, (b) 108, (c) 86, (d) 69, (e) 54, and (f) $30\mu \mathrm{J}$.

Fig. 6

Microscopic image of the retina (a) before and (b) after peeling neural retinal layers off. (c) The energy level of treated spots.

Average peak values of svOCT depend on the laser pulse energy; the result is summarized in Fig. 7(a). The data with blue circles indicate when the laser pulses induce a lesion in the upper neural layers in addition to inducing a lesion in the RPE layer [corresponding to the indicated spots in Fig. 6(a)]. The data represented by orange squares represent the cases when the lesions were induced only on the RPE layer, indicated by white triangle arrows in Fig. 6(b). The laser-induced lesion confined only to the RPE layer can be considered as a successful treatment. Fifteen peaks for each spot are averaged, and the standard deviation is shown by the error bar. As expected, it was difficult to define the threshold energy level that induces lesion only in the RPE layer. If the null hypothesis is defined as a successful treatment and the treatment is decided to be successful when the energy level is in the range from 41.9 to $92.1\mu \mathrm{J}$, which is determined by logistic regression, the type I error and the type II error were 26.7% and 25%, respectively. Compared with the energy level, the average peak values of svOCT showed a better correlation with the lesion creation. Average sv peak values ranged from 12.4 to 38.7 when the induced lesions were observed in both neural and RPE layers; sv values ranged from 0.3 to 18.3 when the lesion was confined only to the RPE layer. In the range of sv peak values from 0.7 to 1.9, no induced lesion was observed at all. When the treatment is decided to be successful with the average svOCT peak value between 1.88 and 15.3 based on logistic regression, the type I error and the type II error were 20% and 0.083%, which was better than the case when the threshold was set by the energy level. For dosimetry control, the method is designed to be used with a power ramped pulsed mode, in which laser energy increases linearly from pulse to pulse within one pulse train and automatically stops the next pulse irradiation when the svOCT peak values reached the predetermined threshold value.

Fig. 7

(a) Averaged peak values depending on pulse laser energy and damage range. (b) Simulated (lines) and estimated temperature from the svOCT intensity (shapes) at neural retina and RPE. (c) Simulated temperature at the neural retina and the RPE as a function of laser energy level for three pulse durations—2, 5, and $10\mu \mathrm{s}$.

In addition, the average peak value of svOCT intensities was correlated to the simulated temperature of the neural retina and melanosome surface in RPE. The linear regression of average peak values of svOCT on laser energy was calculated for photocoagulated lesions and selectively damaged lesions. Then, each of them was correlated to the simulated temperature of neural retina and RPE because the tissue damage process was different from each other. Since the simulated temperature of melanosome surface in the RPE was also linear to pulse laser energy, it was correlated to the average peak values of selectively damaged lesions as

Figure 7(b) shows that the temperature of the neural retina was mostly estimated to increase higher than 50°C when the neural retina was photocoagulated.

We also simulated temperatures of the neural retina and the melanosome surface in RPE as a function of laser energy level for three pulse durations—2, 5, and $10\mu \mathrm{s}$, as shown in Fig. 7(c). The temperature of the neural retina does not change significantly, but the temperature of the melanosome surface in RPE decreases as pulse duration increases. The decrease in temperature can be explained by less heat confinement with longer pulse duration due to heat diffusion during irradiation.

4.

Conclusion

It was shown that the SRT could be successfully monitored by the svOCT imaging system when integrated with the SRT system. We tested our system performance using ex vivo bovine eyes; the svOCT showed distinctive signal variation corresponding to each laser pulse irradiation. The signal variations were proportional to pulse energy levels, and it had a reliable correlation with the creation of lesion within the retina. The temperature at the neural retina and RPE was estimated by svOCT peak values using temperature simulation results, which was consistent with the observed lesion creation. However, we could have missed some minor tissue damages when assessing the photocoagulation and RPE cell damage from the microscopic images. Therefore, more studies that incorporate further analysis supported by histology or fluorescence microscopy would be needed to obtain a more accurate correlation between the svOCT signal and retinal damage range. In addition, we plan to perform in vivo studies using a live animal model to fully validate the utility of this method as an automatic dosimetry control in clinical SRT systems.