Comparative study of retinal injuries induced by a 420-750 nm supercontinuum source and a 532 nm laser

: With the rapid developments and widespread applications of supercontinuum (SC) sources, ocular damage induced by this new light source becomes possible and receives our concern. To explore the ocular damage effect of an SC source, a series of experiments were conducted in a chinchilla grey rabbit model to determine the in-vivo retinal damage thresholds induced by a 420-750 nm SC source and a 532 nm laser. For the SC source, the beam divergence and the corneal 1/e 2 beam diameter were 3.8 mrad and 2.45 mm, respectively. The determined ED 50 values given in terms of total intraocular energy (TIE) for exposure durations of 0.1, 1.0, and 10.0 s were 1.57, 12.1, and 86.0 mJ, respectively. For the 532 nm laser, the beam divergence and the corneal 1/e 2 beam diameter were 0.9 mrad and 2.25 mm, respectively. The determined ED 50 value for an exposure duration of 0.1 s was 1.39 mJ. By employing the retinal thermal action spectrum in the ICNIRP guidelines, the damage thresholds for SC sources could be compared with the exposure limits for incoherent and laser radiation. Between the 420-750 nm SC source and the 532 nm laser, no significant difference could be found for the damage effects including damage threshold, retinal lesion size, and histological damage characteristics.

theoretical and experimental foundations for its ocular protections. In a previous report [20], we determined the rabbit retinal damage threshold induced by a SC source which covered the wavelength spectrum from 450 nm to 2500 nm. For the exposure duration of 0.1 s, the ED 50 damage threshold was about 68.1 mW. This value was significantly higher than the values for lasers in visible wavelengths (11.2-24.7 mW) [17], but obviously lower than the values for infrared wavelengths such as 1064 nm and 1319 nm (145 mW and 13.6 W respectively) [21,22], indicating that the retinal injury induced by the SC source was a combined effect of visible light and near infrared (IRA) radiation. As we know, the retinal minimum spot size for visible radiation is in the range of 7-30 μm, and enlarges to 30-110 μm in the near-infrared wavelength range due to dispersion effect [17]. To reduce the influence of dispersion effect on damage threshold, it is necessary to select the visible spectrum in SC source to investigate the retinal injury effects. As such, a "worst-case" exposure condition would be simulated and lowest damage thresholds could be obtained. In this paper, a comparative study of retinal injuries induced by a 420-750 nm SC source and a 532 nm continuous wave laser was performed in the chinchilla grey rabbit model to explore the damage characteristics and determine the damage thresholds of SC source. The results may contribute to the knowledge base for hazard evaluations of SC source. Figure 1 shows the experimental set-up for retinal exposures by SC source. The SC source (National University of Defense Technology, Changsha, China) produced about 1.4-ns pulses at a repetition rate of 300 kHz and covered the wavelength range from 420 nm to 2500 nm. The output power of the SC source was about 1.5 W, with power stability within ± 2% for 4 hours operating time. The spectrum of the SC source was measured by a spectrograph (OL750, Optronic Laboratories, America) and is shown by the red dashed line in Fig. 2(A). Using a short-pass filter (FESH0750, Thorlabs, America), the visible spectrum from 420 nm to 750 nm was preserved for retinal exposures, with the power of 22 mW. The transmission spectrum of the short-pass filter is shown by the red dashed line in Fig. 2(B) and the spectrum after the filter is shown by the blue solid line in Fig. 2(A). Some K9 glass plates were employed to attenuate the power incident on the animal cornea. The transmittance of the attenuators was measured by a spectrophotometer (Cary 5000, Agilent, America) and kept nearly constant as 0.915 in the wavelength range of 420-750 nm. An electronically-controlled mechanical shutter was employed to control the exposure duration for the SC source. The exposure durations selected for this experiment were 0.1, 1.0 and 10.0 s. A low-power 655 nm laser pointer, coaxial with the SC source, was used as a pointer for exposure position. The transmission spectrum of the reflector was same with the attenuators. A power meter (3A, Ophir, Jerusalem, Israel) was placed at animal eye's position to measure the power incident on the cornea. The 1/e 2 divergence of the SC source was about 3.9 mrad, measured according to the ISO 11146 standard [23]. The irradiance of the SC source was nearly Gaussian distributed which was characterized through the knife edge method [24]. The 1/e 2 diameter of the spot on the cornea was about 2.45 mm. Figure 3 shows the experimental set-up for retinal exposures by 532 nm laser. The power of the laser (Viasho, Beijing, China) could be continuously adjusted through the drive current, having a maximum output of 500 mW. A constant proportion of the laser power was reflected by the prism onto a laser power meter (3A, Ophir, Jerusalem, Israel) for monitoring the stability of the laser power. The prism was employed to change the laser direction for the purpose of keeping the damages at the posterior-pole retinal region and away from the optic disc. A power meter (3A, Ophir, Jerusalem, Israel) was placed at animal eye's position to measure the power incident on the cornea. The divergence of the laser was 0.9 mrad and the 1/e 2 diameter of the spot on the cornea was about 2.25 mm, measured according to the same methods for the SC source.

Animal subjects
Chinchilla grey rabbits of either sex were selected for retinal exposures. The total number was 24. All animals, weighing 2.4-3.0 kg, were procured and maintained in the Center for Laboratory Animal Medicine and Care, Academy of Military Medical Sciences, Beijing, China and used in accordance with the institutional guidelines of the Animal Care and Use Committee; and the ARVO Resolution on the Use of Animals in Research. Subjects were preexposure examined by a slit-lamp (Topcon, Tokyo, Japan) and a fundus camera (Vet 2, Optomed, Finland) to insure clear refractive media and normal fundus. To ensure the subjects did not experience pain and distress, all rabbits were anesthetized with an intramuscular injection of a mixture of ketamine hydrochloride (40 mg/kg) and xylazine (12 mg/kg). Full pupil dilation was performed with two drops of proparacaine hydrochloride 0.5%, phenylephrine hydrochloride 2.5% and tropicamide 1% at a 5-minute interval. The anesthetized animals were placed in a turntable where they were positioned with the aid of the low power laser pointer. The cornea was kept hydrated with one drop of physiological saline solution about 10 s prior to exposure and the excess fluid was blotted at the limbus.

Experimental procedures and data analyses
Using 4 rabbits, a pilot experiment was firstly conducted to estimate the approximate range of the damage thresholds for selected exposure durations (0.1, 1.0, and 10.0 s for SC source, and 0.1 s for 532 nm laser). From the post-exposure readings, we estimated the approximate range of the damage threshold for each condition. Based on the estimations, four different power levels were chosen in the subsequent experiment to determine the damage threshold, as shown by the left two columns in Table 1 and Table 2. Overlapping of retinal lesions was avoided by changing the incident beam angle.
Following each exposure session, lesion/no lesion determinations were made by three experienced investigators for each exposure site at 1, 24, and 48 hours post-exposure, with at least two of three in agreement to confirm a positive reading. Considering the fact that the retinal lesions were most distinguishable at 24-h post exposure, the lesion/no lesion data were collected at this time point. The SAS statistical package (Version 6.12, SAS Institute, Inc., Cary, NC) was employed to analyze the damage data. Specifically, using the collected damage probabilities at different laser power levels, the dose response curve could be obtained, and then the widely accepted Bliss probit analysis was performed to determine the ED 50 thresholds, fiducial limits at the 95% confidence level and probit slopes S (ED 84 /ED 50 ) [25]. The ED 50 refers to the effective dose corresponding to 50% damage probability [25].
A fundus camera (Vet 2, Optomed, Finland) was employed to capture the retinal damage pictures. Furthermore, histopathologic studies were also performed. Some rabbits were euthanized at 24 hours post-exposure. After the euthanization, the eyeballs were taken and fixed in Davidson solution for 24 hours, then dehydrated with ethanol, embedded with paraffin, serially sectioned, and the sections stained with hematoxylin and eosin (H&E). A microscope (BX43F, Olympus, Tokyo, Japan) was used to observe and capture the damage pictures.
For comparing the damage thresholds induced by SC with exposure limits in ICNIRP guidelines [18,19], a theoretical method was developed to calculate the effective exposure dosage. Owing to the similar characteristics between SC source and laser (high spectral brightness and good direction), it is appropriate to use the total intraocular energy Q (TIE, expressed in J) to express the damage threshold. On the other hand, SC source has another unique characteristic, i.e. super broadband width, thus we introduced the retinal thermal hazard spectral weighting function R(λ) [18] (shown in Fig. 4) which characterizes the spectral efficiency to cause threshold retinal thermal damage and has been used for evaluating the hazard risk of incoherent light source. In this way, the effective TIE Q eff could be defined to describe the exposure dosage of SC source. The power of a SC source incident on the cornea can be expressed as is the laser energy incident on the cornea. In ICNIRP guidelines for laser radiation [19], exposure limits for thermal retinal injury in the wavelength range 400-1400 nm are governed by a combined correction factor C A (λ)·C C (λ) which has exactly the same background as the R(λ) function of the broadband guidelines. The spectral correction factor C A (λ) approximates the reciprocal of the RPE absorbance and is defined for 400 nm < λ < 1400 nm; the spectral correction factor C C (λ) approximates the reciprocal of the spectral transmittance of the pre-retinal ocular media and is defined for 700 nm < λ < 1400 nm. The factor (C A (λ)·C C (λ)) −1 is plotted in Fig. 4 and can also be employed to calculate the effective power P eff of SC source.  Table 1 and Table 2 show the experimental conditions and the corresponding damage probabilities at 24-h post-exposure for the SC source and the 532 nm laser, respectively. The damage thresholds expressed in TIE (Total intraocular energy), the fiducial limits at the 95% confidence level and the probit slopes (ED 84 /ED 50 ) are analyzed and listed in Table 3. Combining the 420-750 nm SC spectrum in Fig. 2 with the retinal thermal hazard spectral weighting function R(λ) or the factor (C A (λ)·C C (λ)) −1 in Fig. 4, the damage thresholds expressed as the effective TIE Q eff are calculated and summarized in Table 3.  Under the fundus camera, the damages induced by the 420-750 nm SC source appeared as small circular gray lesions at 1 hour post-exposure. At 24 hours post-exposure, some black dots appeared around the damage lesions (some lesions could not be captured by our fundus camera, but could be clearly observed with eyes through another Topcon fundus camera which lost the photograph function), and lesions which were difficult to distinguish at 1 hour post-exposure became distinct. Therefore, the damage thresholds were determined at 24 hours post-exposure. The appearance of the lesions induced by the 532 nm laser was similar to that of the SC source. We also noted that no obvious difference for the lesion size existed between the SC source and the 532 nm laser. Figure 6 shows the light micrographs of lesions fixed at 24 hours post-exposure. For the picture (A) corresponding to the 420-750 nm SC source irradiation, the exposure duration was 0.1 s and the incident power was 20.1 mW (1.28 times ED 50 value). For the picture (B) corresponding to the 532 nm laser irradiation, the exposure duration was 0.1 s and the incident power was 18.0 mW (1.29 times ED 50 value). The damage characteristics for these two light sources were similar. Typical characteristics included the followings. Cells of the outer nuclear layer were unorderly arranged and some nuclei were heavily stained. The tips of photoreceptor outer segments adhered to the apical portions of the RPE cells. No obvious changes were found in the inner nuclear layer, inner plexiform layer, and ganglion cell layer.

Discussio
Eye is the mo [26]. To pr  [18,19] to retina results from temperature elevation in RPE for exposure durations less than 10 s in the wavelength band of 400 to 1400 nm. Thermo-acoustic damage occurs for exposure durations less than about 10 μs. Optical breakdown and plasma formation dominate from subnanosecond exposures. In our experiment, the SC source did not have CW emission but emitted nanosecond pulses with a high repetition rate of 300 kHz. Existing studies have shown that the injury mechanism in the nanosecond time regime is micro-cavitation and the damage thresholds are significantly lower than the values for thermal damage mechanism [27,28]. But for short pulse lasers with a high repetition rate, the retinal damage mechanism is thermal. For example, Thomas et al. compared the retinal effects from continuous wave and femtosecond mode-locked (120 femtoseconds with 76 MHz repetition rate) lasers in the rhesus model [29]. They found that the determined ED 50 values for 0.25 s exposure duration were 5.9 mJ and 5.84 mJ for mode-locked and CW exposures respectively. The nearly identical damage thresholds indicated a thermal tissue damage mechanism for the femtosecond mode-locked lasers with high repetition rate. Our results also showed that the damage threshold for SC source with 0.1 s exposure duration was slightly higher than the value from CW 532 nm laser, indicating that the damage mechanism was thermal. The typical histological characteristics were similar for thermal damages induced by SC source and CW 532 nm laser, and the main damage targets were photoreceptor and RPE. For SC source, the key question is that which kind of exposure guideline is more appropriate for hazard assessments. In order to answer this question, we first need to understand the basic characteristics of this new light source. The significant characteristics of SC source include super broad bandwidth, high spectral brightness, high spatial coherence, and low beam divergence [7]. In comparisons, lasers can be designed to have an extreme bight output with diffraction limit, but output with broad bandwidth like SC source has not been obtained at present. While for traditional incoherent light source, broad visible and infrared radiation can be generated, with low brightness and poor directionality. As seen in Fig. 5, retinal lesion sizes induced by SC source were nearly same with the sizes for the 532 nm laser. Considering the similar damage effects between SC source and laser, existing damage thresholds induced by lasers should be collected to analyze the values determined in this paper. For comparing the SC damage thresholds with existing exposure limits, the retinal thermal hazard function R(λ) in the ICNIRP guidelines for incoherent radiation could directly be employed to normalize the damage data. However, considering the high spatial coherence of SC source and the resulting minimum retinal spot size similar with lasers, the laser guidelines may be more appropriate to evaluate the potential hazard induced by SC source. For the case of multiple wavelengths, the ICNIPR laser guidelines describe that the respective exposures have to be treated as additive when the same kind of tissue is at risk and are treated independently when different tissues are at risk. In the book by Henderson et al. [26], the evaluation method for broadband exposure is presented. By weighting the data of spectrum (the exposure level) by an action spectrum and summing up the wavelength range, the effective exposure level could be obtained and compared with the minimum laser exposure limit. For the retinal thermal damage in the 400-1400 nm wavelength band, the reciprocal of C A (λ)·C C (λ) is nearly equivalent to the action spectrum R(λ) which is defined for broadband incoherent exposure limits, although some minor deviations at the blue edge and near-infrared edge exist as shown by Fig. 4. One fact should be noted that the action spectrum is a simplification, as the injury thresholds reflect some level of wavelength dependence within 435 nm and 700 nm where R(λ) equals unity. This wavelength dependence of damage thresholds has been discussed by Lund et al. through analysis of experimental values [30,31] and by Schulmeister et al. through a computer model [32]. Therefore there would be an option to make the action spectrum more accurate but the disadvantage is an increased complexity of guidelines. According to above analysis, both the retinal hazard function R(λ) in ICNIRP guidelines for incoherent radiation and the reciprocal of C A (λ)·C C (λ) in ICNIRP guidelines for laser radiation could be employed to calculate the effective exposure of SC source, as described in the "Methods" section. In the followings, we firstly compared existing rhesus injury data with our values in rabbit model, and then discussed the applicability of the available exposure limits.
Lund et al. summarized the rhesus retinal damage thresholds for the exposure duration of 0.1 s [30]. For lasers in the wavelength band from 420 to 750 nm, the ED 50 values lie from 1.12 to 2.24 mJ. The ED 50 value for 532.0 nm emission line is 1.42 mJ, and the value for 750.0 nm wavelength is 2.24 mJ. Above values are all determined for extra-macular retina. In another report from Lund et al. [33], the determined macular ED 50 for 0.1 s exposure duration and 514 nm laser radiation is 1.05 mJ for the minimum retinal image size at 24 h postexposure, and the extra-macular value under the same conditions is 1.23 mJ. In our study, the rabbit retinal damage threshold for 532 nm laser was 1.39 mJ which was some factor above the macular ED 50 from modern good rhesus studies but was quite similar with the ED 50 values for extra-macular retina. For the SC source, the ED 50 value expressed in TIE was slightly larger than the value for 532 nm (1.57 mJ vs.1.39 mJ), which possibly resulted from the higher proportion of red light in the SC spectrum. As shown by Lund et al [30], the retinal damage thresholds for wavelengths between 600 and 750 nm are systematically larger than values for wavelength band from 420 to 600 nm, probably due to the lower absorption of retina for red light than that for green light. Thus, the overall effect would be a slight higher ED 50 value for the SC source than that for the 532 nm laser.  [36]. The ED 50 values varied from 56 mJ to 170 mJ as the 532/860 power ratios decreased from 7/1 to 0/1. For comparison, the determined ED 50 value for the 420-750 nm SC source was 86.0 mJ, which fell in the range of values by Roach et al [36]. Here, one important fact to point out is that all the above mentioned damage values were determined with small retinal irradiance diameters (< 100 μm). As we know, retinal damage threshold ED 50 , expressed as TIE, is dependent on the retinal image size [37,38], and should be obtained under controlled conditions to simulate "worst-case" exposure conditions including that the incident beam is collimated. According to Schulmeister et al [37,38], the damage thresholds, expressed as TIE, kept nearly constant for exposure duration of 0.1 s as the retinal spot diameter was smaller than about 100 μm. In our experiment, the beam divergence of the SC source was about 3.9 mrad and the corresponding retinal spot diameter was about 37.6 μm (the focal length for rabbit is 9.9 mm for a relaxed eye). Therefore, we compared our data with previous ED 50 values with small image sizes.
Exposure levels lower than ELs (Exposure limits) are not expected to cause adverse biological effects. ELs are established based on damage thresholds determined in appropriate animal models. There exists a safety margin between ED 50 and the corresponding EL, and the safety margin is defined as the ED 50 /EL ratio which is generally preferred to be about 10 for retina. Figure 7 shows the rabbit retinal ED 50 damage thresholds (expressed in Q eff ) induced by the 420-750 nm SC source and the 532 nm laser. The ELs for incoherent radiation and laser radiation were also included [18,19]. In the ICNIRP guidelines for incoherent radiation, the "normal" retinal thermal exposure limits are based on pupil constriction following exposure to bright light and the EL is a constant radiance for exposure durations above 0.25 s. But the guidelines also note that a more conservative EL (dependent on exposure duration) should be applied for the case of weak visual stimulus. Apparently, the latter was more suitable for the evaluation because the animals in our experiment were anaesthetized with fully dilated pupils. The EL for laser radiation is the minimum exposure limit in the 400-700 nm wavelength range. Some conclusions, indicated by Fig. 7, were as followings. Both the exposure limits for incoherent and laser radiation better reflected the dependence on exposure duration as in the damage thresholds. The ratios between ED 50 s and ELs for laser radiation were 12.3, 16.9 and 21.4 respectively for exposure durations of 0.1, 1.0 and 10.0 s. The EL for incoherent radiation is a factor of about 1.26 higher than the equivalent laser limit, indicating that the laser limits are more conservative as compared to the damage thresholds. Fig. 7. Black square points: rabbit retinal ED 50 damage thresholds induced by 420-750 SC source exposure. Green circle point: rabbit retinal ED 50 damage threshold induced by 532 nm laser. Blue solid line: retinal thermal exposure limit for incoherent radiation (small sources) and the radiance in the guidelines has been transformed to intraocular energy by assuming a 7 mm pupil. Exposure durations were from 0.1 s to 10.0 s. Red dashed line: retinal thermal exposure limit for laser radiation in the wavelength range of 400-700 nm by assuming a 7 mm pupil.

Conclusion
A comparative study of retinal injuries induced by a 420-750 nm supercontinuum source and a CW 532 nm laser were performed in the chinchilla grey rabbit model. The SC source emitted about 1.4-ns pulses at a repetition rate of 300 kHz. For exposure durations of 0.1, 1.0 and 10.0 s, the retinal damage thresholds were 1.57, 12.1, and 86.0 mJ respectively. For the 532 nm laser, the retinal damage threshold was 1.39 mJ with the exposure duration of 0.1 s. No significant difference could be found for the damage effects between the SC source and 532 nm laser, indicating that the damage mechanism of the SC was thermal. Damage threshold for rabbit was quite similar with the available rhesus extra-macula data, supporting that the rabbit model can be characterized as a valuable model for determining injury thresholds for laser and optical radiation particularly when damage trends are investigated. Results confirmed that the effects were additive for retinal exposures to multiple wavelengths or broadband light source. By employing the action spectrum for retinal thermal damage, the exposure for SC could be normalized and then compared with exposure limits in ICNIRP guidelines. Both the exposure limits for laser radiation and broadband incoherent radiation reflected the dependence of exposure duration as in the SC damage thresholds. Enough safety margins of 12.3~21.4 existed between the thresholds and the exposure limits for laser radiation.