Evaluation of the potential eye hazard at visible wavelengths of the supercontinuum generated by an ultrafast NIR laser in water

: Lasers with ultrashort pulse durations have become ubiquitous in various applications, including ocular surgery. Therefore, we need to consider the role of nonlinear optical effects, such as supercontinuum generation during propagation within the ocular media, when evaluating their potential hazard. We used a NIR femtosecond laser to generate a supercontinuum within an artificial eye. We recorded the visible spectra of the supercontinuum generated and calculated the energy contained within the visible band. Our results indicate that for wavelengths between 1350 nm and 1450 nm the energy contained within the visible band of the generated white light supercontinuum may surpass current safety exposure limits, and pose a risk of injury to the retina.


Introduction
In recent years, pulsed lasers with ultrashort pulse durations, i.e. shorter than 10 ps, have become ubiquitous in various applications [1][2][3][4][5] including medical procedures such as laser eye surgery [6][7][8]. The peak power of these sources is sufficiently high that the role of nonlinear optical effects resulting from the interaction of ultrashort pulses with the surrounding media needs to be considered when evaluating their potential hazard to the retina [9,10].
Ocular laser safety standards define the maximum permissible exposure (MPE) that can enter the eye for a specific wavelength, beam size, and exposure duration [11][12][13]. Current laser safety standards, such as the American National Standard for Safe Use of Lasers (ANSI Z136.1-2014) [11], the International Electrotechnical Commission for safety of laser products (IEC 60825-1:2014) [12], and the International Commission for Non-Ionizing Radiation Protection (ICNIRP) guidelines [13], include MPE limits for wavelengths ranging from 400 nm up to 1400 nm with pulse durations longer than 100 fs. The majority of the retinal threshold studies used to establish the current MPEs have used data from in vivo studies on non-human primates [14][15][16][17]. For wavelengths longer than 1400 nm and pulse durations shorter than 1 ns, no exposure limits are provided due to lack of biological data, but the recommendation is to use the MPE applicable to 1 ns pulse durations [11]. These standards tacitly assume that there is only linear optical propagation of light within the ocular media; therefore, the wavelength of the incoming laser beam remains unchanged as it propagates through the eye. However, there have been various studies demonstrating nonlinear optical effects in water resulting from the interaction with ultrashort pulses which lead to the generation of wavelengths other than the incoming wavelength [18][19][20][21][22][23][24][25]. One such nonlinear effect is the generation of supercontinuum (SC) light, i.e. ultra-broadband radiation with a high power spectral density and a high degree of spatial coherence [26]. The precise physical processes involved in producing the spectral broadening of the incoming laser pulses depend on the specific refractive index dispersion in the material as well as pulse duration, power, and wavelength of the incident beam [27]. In particular, for femtosecond pulse durations in water, self-phase modulation, self-focusing and solitons dominate the generation of the SC [19,27,28]. The interplay between the different nonlinear effects involved results in a threshold pulse energy for the incoming beam, above which SC generation is observed [28][29][30]. Given that the generated SC travels in the same direction as the incoming beam [27], and the main component of the vitreous humor is water, these studies underscore the need to evaluate whether or not the SC generated by femtosecond pulses in ocular media poses a hazard to the human eye, specifically, the retina [23][24][25].
The near infrared (NIR) region of the electromagnetic spectrum is of particular interest from a laser safety standpoint because it includes a region of strong water absorption above 1064 nm where light propagates in an anomalous dispersion regime [19,22,31]. In addition, as the wavelength increases from 1200 to 1400 nm, due to the optical properties and absorption characteristics of the tissues, there is a gradual transition in which tissue is most sensitive to laser damage. For wavelengths in the visible range and up to ∼1064 nm the damage threshold for the retina is less than that for the cornea, however, above 1400 nm the damage threshold for the cornea dominates [32][33][34] defining the transitional NIR region. In this study, we used a NIR femtosecond laser at wavelengths between 800 nm and 1500 nm to determine the threshold energy needed to generate a SC within a water-filled artificial eye. The artificial eye was designed to emulate the path length and focusing geometry of a non-human primate eye and was inspired by previous studies [35]. By comparing the SC generation threshold energies with the laser safety limits at NIR wavelengths, we identified wavelengths with incident energies below the laser safety limits. We recorded the visible spectra of the SC generated by the NIR wavelengths of interest at two different pulse energies and, by comparing with the spectra of a known tungsten-halogen source, calculated the energy contained within the visible band. Subsequently, we compared the SC energy generated within the visible band to the maximum permissible energy derived from the ANSI Z136.1-2014 MPE limits for visible wavelengths in order to determine if there was a potential for the corresponding exposure conditions to cause damage to the eye. These results increase our current understanding of nonlinear optical hazards associated with NIR ultrashort pulsed lasers, guide future in vivo experiments to establish ocular damage thresholds from SC generated by NIR ultrafast lasers, and contribute to the knowledge base used for setting laser safety standards in the transitional NIR region for ocular exposure to ultrashort pulses.

Optical setup for supercontinuum generation
An optical parametric amplifier (OPA; HE-TOPAS, Light Conversion Technologies) pumped by a Ti:Sapphire regenerative amplifier (Spitfire Ace, Spectra Physics) was used to produce excitation wavelengths between 800 nm and 1500 nm with a 1 kHz repetition rate and a nominal pulse duration of 100 fs. As shown in Fig. 1, we sent the output from the OPA through a set of neutral density filters to control the pulse energy and a telescope to reduce the approximately 8 mm diameter beam to a collimated 3 to 4 mm diameter beam. For wavelengths above 900 nm, we used an 850 nm long-pass filter to block any residual light from the 800 nm pump beam and any visible light components produced by the NIR light generation in the OPA. The collimated NIR light at a 1 kHz repetition rate was directed into an artificial eye that consisted of a focusing lens and a flat window at opposite ends of a water-filled container as detailed in section 2.1.2.
The SC light generated exited through the back window of the artificial eye and was collected by a Spectralon coated, 38.1 mm diameter integrating sphere with a 9.5 mm sample port opening (FOIS-1 Integrating Sphere, Ocean Insight). Using an optical fiber, we sent the output of the integrating sphere to a visible light spectrometer (USB 2000, Ocean Optics) or a NIR spectrometer (NIR 512, Ocean Optics) in order to capture the SC spectra in the visible and IR ranges separately. The spectra were continuously collected for specific integration times and were averaged by the spectrometer's software. For some measurements, we placed a short-pass filter (FF01-890/SP-25, Semrock) before the integrating sphere. This filter passed wavelengths between 370 nm and 875 nm, essentially blocking any UV and NIR components present in the light exiting the artificial eye and enabling us to isolate the contribution from the visible range, i.e. 380 to 740 nm, through our subsequent numerical analysis of the SC spectra. We used a power meter to measure the energy of the incoming NIR beam (PM10, Coherent) and an energy meter (J-10MB-LE, Coherent) or a power meter (919P-003-10, Ophir) to obtain the energy contained in the outgoing SC light depending on its magnitude. We built an enclosure with an opening for the NIR beam around the artificial eye and the integrating sphere to prevent ambient room light from being collected during spectral and energy measurements as determined by the suppression of the fluorescent light peaks in the visible spectrum.
We characterized the incoming NIR laser beam by measuring the IR spectrum with the NIR spectrometer and collecting the beam profile using an infrared camera (Pyrocam-III, Spiricon) right before the artificial eye. For all wavelengths used, the beam was approximately Gaussian. We calculated the diameter of the incident beam at the 1/e 2 intensity point by averaging two separate knife-edge measurements 19 cm before and 19 cm after the artificial eye's position along the optical path when no artificial eye was present. The diameter of the beam depended on the wavelength and on the precise alignment of the telescope, varying between 2.65 mm at 1350 nm and 4.14 mm at 1400 nm.

Artificial eye design
We designed the artificial eye to emulate the optical path length and focusing geometry of a living eye, specifically that of a non-human primate, while allowing access to light traveling to the back of the eye. It consisted of a focusing lens and a flat exit window at opposite ends of a 30 mm x 50 mm x 55 mm tall container filled with distilled water as shown in Fig. 2(a). The eye of a non-human primate has an effective focal length of 17 mm in water [36] (13.5 mm in air [37]) so the lens was selected to have a similar focal length in water which resulted in a nominal back focal length of 17.98 mm at 800 nm and 18.32 mm at 1450 nm. The container was 3D printed out of Tough PLA (Ultimaker S5, Ultimaker B.V.) before threading the two openings on opposing sides to hold the optical components, see Fig. 2 The opening for the incoming NIR beam held an 8 mm diameter plano convex lens, f = 15 mm (SLB-08-15P, Opto Sigma) mounted onto a lens holder adaptor (LMRA8, Thorlabs) within a 12.7 mm diameter adjustable lens tube (SM05V10, Thorlabs), assembly I in Fig. 2(b). The opening for the outgoing generated SC held a 25.4 mm diameter lens tube cap (SM1CP2M, Thorlabs) machined to mount the 12.7 mm diameter, 1 mm thick Infrasil exit window (UDP05, Thorlabs), labeled as assembly II in Fig. 2(b). The exit window had a relatively flat spectral response and high transmission at both visible and IR wavelengths. The artificial eye was mounted on a multi-directional positioning stage (TTR001, Thorlabs) in order to facilitate the alignment of its optical axis to the laser path. The plano convex lens focused the incoming NIR light to a point within the water-filled container; furthermore, the artificial eye's design allowed for adjustments of the focal plane relative to the exit window by changing the amount the adjustable lens tube protruded into the artificial eye's chamber. The distance between the back plane of the lens and the front plane of the exit window was set at the beginning of the study and remained fixed at approximately 24 mm, leaving about 6 mm between the focal point and the front face of the exit window. Note that the exact separation depended on the wavelength of the incoming beam due to chromatic focal shift. Figure 2(c) is an optical image (real-color) of the artificial eye showing the SC generated at an incident wavelength of 1300 nm as a white-blueish light within the aperture of the power meter approximately centered in the upper half of the picture. The red and white arrows indicate the beam paths of the incoming NIR ultrafast laser pulses and the generated SC respectively.

Determination of the threshold energy for supercontinuum generation
In order to find the threshold energy (E Th ) for SC generation at each NIR wavelength selected, we systematically varied the power of the incoming beam with a set of neutral density filters while monitoring the SC spectra with the visible spectrometer. Once we could resolve a clear spectrum above the noise floor, we obtained the incoming pulse energy by placing a power meter in front of the artificial eye and dividing the measured (average) power by the repetition rate of the laser and labeled it the E Th . Subsequently, we removed the power meter and recorded the visible spectrum directly, the visible spectrum with the short-pass filter before the integrating sphere, and the IR spectrum. For each spectrum measured, we collected the corresponding dark spectrum by blocking the incoming NIR beam with a mechanical shutter. We then replaced the integrating sphere with an energy meter and recorded the energy in the generated SC with the short-pass filter in place. Note that at some incoming wavelengths, we could visually distinguish SC being generated at energies lower than the nominal threshold energy but were unable to measure a spectrum because it was below the sensitivity of our detection system. Similarly, for some incoming wavelengths, the energy of the generated SC with the short-pass filter in place was below the detection limit of our detectors although we could record a spectrum.

Quantification of the supercontinuum's energy contained in the visible range
In our study, the detection system consisted of the exit window of the artificial eye, the short-pass filter (if in place), the integrating sphere, the optical fiber, and the visible spectrometer. The wavelength dependent response of the detection system altered all measured spectra. To correct for this effect, we followed standard procedures used in the spectroscopy and radiometry communities [38][39][40][41]. Briefly, these procedures consist of utilizing a known broadband source to obtain the instrument response function (IRF) and then using the IRF to correct the shape and amplitude of the measured spectra. In our case, we used a tungsten-halogen broadband source (LS-1, Ocean Optics) and we obtained the IRF of the broadband source-exit window-integrating sphere-optical fiber-spectrometer combination. Subsequently, we corrected the SC spectrum collected with the visible spectrometer for the effect of the detection system as follows: where SC was the spectrum of the SC corrected for the transmission spectrum of the exit window and includes the respective dark spectrum, i.e. the spectrum collected with the light source blocked, and the IRF was obtained following standard procedures. The energy of the SC generated within the visible range was obtained from where the energy calibration factor (ε) is a scalar quantity that relates the intensity counts of the visible spectrometer to the energy contained in the spectrum of a known source, in this case the tungsten-halogen lamp, and the integration limits λ 1 =380 nm and λ 2 =740 nm correspond to the visible range of the electromagnetic spectrum. The energy calibration factor was given by where the lamp's output energy was measured with the short-pass filter in place in Joules and the denominator was in counts. The integrand of the denominator (corrS F ) represents the scaled shape-corrected source's spectrum measured with the short-pass filter in place and includes the corresponding dark spectrum while the integration limits λ 3 =370 nm and λ 4 =875 nm correspond to the wavelength range of the short-pass filter. Note that the wavelength range of the filter encompasses the visible range, reducing any ambiguities introduced by having contributions from wavelengths outside the range of interest when calculating the energy in the visible range of the generated SC. All numerical integrations were calculated in Matlab and utilize the trapezoidal method with unit spacing.  [11]. Given that we have nominally 100 fs long pulses at a repetition rate of 1 kHz, we used the MPE NIR values for single-pulse exposures per guidance. The last column of Table 1 lists the ratio of the threshold pulse energy to the maximum permissible energy where a ratio of less than 1 means that the incident pulse energy needed to generate SC light for that wavelength was lower than the corresponding TIE NIR , therefore that incident pulse energy is considered safe per current laser safety standards. This ratio was used to determine the wavelengths of interest to be examined further in order to determine if the generated SC contained enough energy at visible wavelengths to potentially cause damage to the retina by overcoming the laser safety limits for visible wavelength. We chose to examine wavelength and energy combinations for which the ratio was less than 10, meaning that the incident pulse energy was less than ten times the corresponding maximum permissible energy, which directed our subsequent analysis to 1350 nm and longer wavelengths. At 1475 nm and 1500 nm, we observed relatively weak third harmonic generation and were unable to deliver enough energy to generate a SC. Consequently, we did not include them in any subsequent analysis or discussion.  Table 2 lists suprathreshold pulse energies which were all less than 2.5 times the respective E Th and, except for 1350 nm, would be considered safe given that they were lower than the TIE NIR corresponding to the MPE NIR , i.e. the ratio of the incident pulse energy to the maximum permissible energy was less than 1 (last column in Table 2). These additional energies have been included to enable a more complete discussion of the role that the incident pulse energy has on the spectrum of the SC generated and its potential to cause damage to the retina by overcoming the laser safety limits for visible wavelengths.  Figure 3 shows the energy spectra of the SC generated by the NIR wavelengths of interest. We obtained the spectra from the corrected visible SC spectrum (corrSC) by multiplying Eq. (1) in section 2.2 by the corresponding energy calibration factor. The black lines correspond to the E Th listed in Table 1 and, for 1350 nm and 1400 nm, have been magnified fifty times. The light blue lines correspond to the suprathreshold energies listed in Table 2. The dashed vertical lines indicate the region used to determine the energy within the visible range while the dash-dotted vertical line is located at the third harmonic of the incoming beam's wavelength. The insets show the spectra of the incoming beam (dotted lines) and the IR spectra of the SC generated (solid lines) between 860 nm and 1660 nm. Table 3 lists the position of the maximal peaks in the visible and IR ranges as well as the limits of the spectral bandwidths of the SC generated for the selected exposure conditions, where the shaded rows correspond to suprathreshold energies. In the visible range, the SC spectra were all blue-shifted relative to the wavelength of the incoming beam and had bandwidths spanning between 110 nm and 450 nm. In addition, we  observed a redshift in the peak of the SC spectrum for 1350, 1375, and 1400 nm as the incident pulse energy was increased. At 1350 nm, the amplitude of the spectrum at E Th was small and relatively featureless (black line in Fig. 3(a)). Increasing the incoming pulse energy to 2.5 × E Th resulted in the development of a clear peak 450 times larger in amplitude and an increase in bandwidth of ∼100% relative to the spectra at E Th , while remaining within 50% of the maximum permissible energy (light blue line in Fig. 3(a)). At 1375 nm, the peak amplitude of the SC spectrum increased by a factor of ∼4 while the bandwidth increased ∼100% when increasing the incident pulse energy to 1.7 × E th (Fig. 3(b)). Similarly, at 1400 nm, increasing the incident pulse energy to 1.4 × E Th resulted in an increase of 80 times in the peak amplitude as well as a ∼4 times increase in bandwidth of the SC spectrum (Fig. 3(c)). These data demonstrate a rapid increase in the amplitude, bandwidth and spectral distribution of the generated SC as the incoming pulse energy is increased above E th [42].

Supercontinuum spectra at wavelengths of interest
For 1350, 1375, and 1400 nm, the peaks in the amplitude did not appear to correspond to the third harmonic of the incoming wavelength. However, for E Th at 1375 nm there was a small inflexion point in the spectrum close to the third harmonic ( Fig. 3(b)), which was absent at 1350 nm and 1400 nm. For 1435 nm and 1450 nm, the spectra for third harmonic generation was superimposed on a broad spectral background (Fig. 3(d) and (e)). As the incident pulse energy was increased, the amplitude of the broad spectral background increased, confirming that it originated from the generation of a SC. At 1435 nm, the peak amplitude of the third harmonic generation did not significantly increase when the incoming pulse energy increased to 2 × E Th (Fig. 3(d)). However, at 1450 nm it increased ∼4 times when the energy of the incoming beam increased to 1.7 × E Th (Fig. 3(e)).
In the IR range, all the SC spectra showed significant broadening when compared to the incoming beam, see insets in Fig. 3. The bandwidths of the SC generated varied from about 360 nm for E Th at 1375, 1400, and 1450 nm to a maximum of 545 nm for the suprathreshold energy at 1350 nm. All IR SC spectra had two blue-shifted peaks relative to the incoming wavelength, with the ones for 1435 nm and 1450 nm separated by a deeper trough. The peak occurring at shorter wavelengths in the IR range was located at (1130 ± 6.6) nm for all incoming wavelengths, where the uncertainty was determined by the wavelength resolution of the IR spectrometer.

Comparison of the energy of the generated supercontinuum contained within the visible range and the maximum permissible energy at visible wavelengths
The primary purpose of this study was to determine if the SC generated by NIR wavelengths with incident energies below the laser safety limits contained enough energy in the visible range to overcome the laser safety limits for visible wavelengths. The set of exposure conditions of interest consisted of incident NIR wavelengths with pulse energies below 10 times the maximum permissible energy obtained from laser safety standards and capable of generating a SC. Results are summarized in Table 4 where the first and second columns list the incident NIR wavelengths and pulse energies used. The third column lists the ratio of the incident pulse energy to the TIE NIR corresponding to the MPE NIR from column 5 in Tables 1 and 2, where a value smaller than 1 means that the incoming pulse energy was lower than the corresponding TIE NIR and would be considered safe by the current standard. With the exception of 1350 nm with an incident pulse energy of 575 µJ, all other conditions fell within this category. The shaded rows in Table 4 correspond to incoming suprathreshold pulse energies.  In order to decide if the SC generated by a set of exposure conditions posed a risk to the retina, we first had to determine how much energy was contained within the visible range. We calculated the energy of the SC generated within the visible range from the corresponding spectra using Eq. (2) in section 2.2. Results are listed in Table 4, column 4 where the uncertainty includes a 10% variability in the spectrum of the SC generated as well as the uncertainty in the determination of the IRF and the ε due to the source's variability. To validate these results, we compared them to the corresponding energy measurements of the SC generated with the short-pass filter at the output of the artificial eye and found that they compared favorably (data not shown). Subsequently we compared the energy of the SC generated within the visible range to the maximum permissible energy for visible wavelengths obtained from the MPE values listed in the laser safety standards for a 7 mm diameter measurement aperture. The MPE limit is constant throughout the visible range at 100 nJ/cm 2 therefore the corresponding maximum permissible energy for visible wavelengths is 38.48 nJ. The last column in Table 4 shows the result of this comparison where ratios larger than 1 indicate that the energy of the SC generated contained within the visible range of the electromagnetic spectrum is larger than the maximum permissible energy at visible wavelengths in current laser safety standards. Note that, although the MPE limits we are using are defined at specifics wavelengths, they establish a limit on the maximum permissible energy, or TIE, deposited on the retina that will not cause damage. Given that the SC generated contains a range of wavelengths, in order to avoid any confusion with the usual interpretation of the TIE for laser sources, which assume a single wavelength, we will continue using the term maximum permissible energy for visible wavelengths.
From Table 4, it can be seen that at 1350 nm, when the incoming pulse energy was approximately 2 / 3 of the corresponding TIE NIR , the energy of the SC generated within the visible range was an order of magnitude smaller than the maximum permissible energy for visible wavelengths. However, when the incident pulse energy was 50% larger than the NIR maximum permissible energy, the energy of the SC generated within the visible range was ∼20 times the maximum permissible energy in the visible range. If, for simplicity, we assume a linear relationship between the incoming pulse energy and the energy of the SC generated within the visible range, we can use our results to estimate the energy of the SC generated at the corresponding TIE NIR by interpolation. In particular, at the TIE NIR corresponding to the MPE for 1350 nm, the energy of the SC generated within the visible range would be about 10 times the maximum permissible energy in the visible range. In contrast, at E Th for 1375 nm, the incoming pulse energy was about one tenth of the corresponding TIE NIR while the energy of the SC generated within the visible range was about 2 times the maximum permissible energy in the visible range. Increasing the incident pulse energy by 80%, while still almost 5 times smaller than the NIR maximum permissible energy, resulted in an increase in the energy of the generated SC within the visible range by a factor of 7, which is 12 times the maximum permissible energy in the visible range.
At 1400 nm, the energy of the generated SC within the visible range at the E Th , which was about 300 times smaller than the corresponding TIE NIR , was about 30 times smaller than the maximum permissible energy for visible wavelengths. However, when the incident pulse energy increased by about 40%, although it was still ∼250 times smaller than the NIR maximum permissible energy, the energy of the SC generated within the visible range surpassed the maximum permissible energy allowed for visible wavelengths by more than a factor of 10. For the E Th at 1435 nm, the incident pulse energy was ∼300 times smaller than the corresponding TIE NIR and the generated SC's energy within the visible range was ∼80% of the maximum permissible energy allowed for visible wavelengths. When the incident pulse energy was increased by a factor of 2, although it was still ∼160 times lower than the corresponding TIE NIR , the energy of the generated SC contained within the visible range surpassed the maximum permissible energy in the visible range by ∼40%. In the case of 1450 nm, the energy of the generated SC contained in the visible range at the E Th was around one tenth of the maximum permissible energy allowed for visible wavelengths although the incident pulse energy was ∼140 times smaller than the TIE NIR . When the energy of the incident beam increased by 70%, the energy contained in the visible range of the SC generated was ∼80% of the maximum permissible energy allowed for visible wavelengths.

Discussion
The spectra for the SC generated in this study presented similar characteristics to those found in the literature for comparable incoming wavelengths in water [19][20][21]24]. For 1350, 1375, and 1400 nm incident wavelengths, the visible spectra for the SC generated with the higher incident pulse energies had a distinct peak that did not correspond to the third harmonic of the incoming beam, are all somewhat asymmetric, and had spectral bandwidths of 340, 430, and 400 nm respectively. In contrast, at 1435 nm and 1450 nm, the visible spectra had a sharp peak that corresponds to the third harmonic superimposed on a broad spectral background that we attribute to SC generation. The cause of these differences in the spectral distribution may be the increase in linear absorption of water for wavelengths above 1400 nm and the relatively long water path for these experiments; ∼18 mm for the incident NIR beam and ∼6 mm for the generated SC. Our results were also in agreement with previous reports where a relatively modest increase in the incoming beam's energy results in a large increase in the amplitude, bandwidth and spectral distribution of the SC generated [24,43].
The primary goal of this study was to determine if the SC generated by NIR wavelengths with incident energies below the laser safety limits contain enough energy in the visible range to overcome the laser safety limits for visible wavelengths. With the exception of 1350 nm with an incident pulse energy of 575 µJ, all the incoming pulse energies for the wavelengths of interest were lower than the TIE NIR corresponding to the MPE NIR and would be considered safe according to current laser safety standards. More so for incoming wavelengths above 1375 nm where all of the incoming pulse energies were at least ∼100 times smaller than the corresponding NIR maximum permissible energy. However, our results for suprathreshold energies showed that the energy of the SC generated within the visible range, surpassed the maximum permissible energy allowed for visible wavelengths by at least a factor of 10 for 1350, 1375, and 1400 nm while the incident pulse energy remained within 50% of the corresponding TIE NIR . For 1435 nm, when the incident pulse energy was 0.6% of the NIR maximum permissible energy, the energy of the SC generated contained within the visible range surpassed the maximum permissible energy in the visible range by 40%. On the other hand, when the incident pulse energy was 1% of the NIR maximum permissible energy at 1450 nm, the highest energy available with our laser system, the energy contained in the visible range of the SC generated was ∼20% below the maximum permissible energy for the visible range. Therefore, for all wavelengths of interest except 1450 nm, the energy contained within the visible range of the generated SC surpassed the maximum permissible energy allowed for visible wavelengths by the laser safety standards for the eye for at least one of the two energy conditions tested. These results agree qualitatively with recent modeling predictions [23,25].
The experiments and analysis presented in this manuscript focused on quantifying the energy of the generated SC contained within the visible range and comparing it to the maximum permissible energy at visible wavelengths obtained from laser safety standards. Laser safety standards define MPE limits for specifics wavelengths; however, the generated SC has components in the NIR in addition to the ones in the visible range that may contribute to the retinal hazard. Furthermore, laser safety standards establish a TIE corresponding to the MPE at a particular wavelength yet the generated SC contains a range of wavelengths all of which contribute to the amount of energy deposited on to the retina. These factors should be considered when designing and interpreting results from in vivo experiments [44]. In addition, our experiments and analysis do not take into account the effects of scattering and Fresnel reflection losses at the surface of the cornea [45,46] on the process of SC generation and, consequently, how they may affect the potential hazard the generated SC presents to the retina. Recently, there have been reports that the addition of salts increases the SC signal relative to that of pure water [47] while the addition of proteins suppresses the SC generation [48,49]. The vitreous humor contains both salts and proteins therefore it is unclear how their apparently competing contributions would impact the SC generation in a more realistic fluid model although the n 2 value for vitreous was measured to be experimentally equivalent to water [50]. In spite of these caveats, our results provide a baseline for pure water and strongly suggest that the SC generated by femtosecond pulses may pose a hazard to the human eye, highlighting the need for further studies.

Summary
In this study, we used a NIR femtosecond laser at wavelengths between 800 nm and 1500 nm to determine the energy threshold for SC generation within a water-filled artificial eye. The artificial eye consisted of a focusing lens and a flat exit window at opposite ends of a container filled with distilled water, emulating the path length and focusing geometry of a non-human primate's eye. We found that the energy threshold for SC generation is only lower than the maximum permissible energy for wavelengths between 1350 nm and 1450 nm, where the linear absorption of water is high. We recorded the visible spectra of the SC generated at the threshold energy and a slightly higher pulse energy. By comparing with the spectra of a known tungsten-halogen source, we calculated the energy contained within the visible band. Our results indicate that for 1350 nm and longer incident wavelengths, the energy of the generated SC contained within the visible range can surpass the maximum permissible energy allowed for visible wavelengths by the laser safety standards, while remaining within the safety standards for NIR wavelengths. Notably, although the incident pulse energy was less than six thousandths of the maximum permissible energy at 1400 nm and 1435 nm, the energy of the generated SC was at least 40% higher than the maximum permissible energy at visible wavelengths. These observations indicate there is a strong potential for some exposure conditions to cause damage to the eye, not necessarily because of the incoming exposure conditions, but because of the SC generated by nonlinear effects in the aqueous media of the eye. Therefore, our results highlight the need to perform in vivo studies to directly evaluate potential damage to the retina from SC generated by ultrashort pulses at 1350 nm and longer wavelengths, and to validate or recommend changes to the laser safety limits for ultrashort pulsed lasers at 1400 nm and longer wavelengths.