Handheld multispectral fluorescence lifetime imaging system for in vivo applications

There is an increasing interest in the application of fluorescence lifetime imaging (FLIM) for medical diagnosis. Central to the clinical translation of FLIM technology is the development of compact and highspeed clinically compatible systems. We present a handheld probe design consisting of a small maneuverable box fitted with a rigid endoscope, capable of continuous lifetime imaging at multiple emission bands simultaneously. The system was characterized using standard fluorescent dyes. The performance was then further demonstrated by imaging a hamster cheek pouch in vivo, and oral mucosa tissue both ex vivo and in vivo, all using safe and permissible exposure levels. Such a design can greatly facilitate the evaluation of FLIM for oral cancer imaging in vivo. ©2014 Optical Society of America OCIS codes: (170.2520) Fluorescence microscopy; (170.3650) Lifetime-based sensing; (170.2150) Endoscopic imaging; (170.3890) Medical optics instrumentation. References and links 1. J. R. Lakowicz, Principles of Fluorescence Spectroscopy (Springer, New York, 2006). 2. J. Mizeret, T. Stepinac, M. Hansroul, A. Studzinski, H. van den Bergh, and G. Wagnieres, “Instrumentation for real-time fluorescence lifetime imaging in endoscopy,” Rev. Sci. Instrum. 70(12), 4689–4701 (1999). 3. I. Munro, J. McGinty, N. Galletly, J. Requejo-Isidro, P. M. P. Lanigan, D. S. Elson, C. Dunsby, M. A. A. Neil, M. J. Lever, G. W. H. Stamp, and P. M. W. French, “Toward the clinical application of time-domain fluorescence lifetime imaging,” J. Biomed. Opt. 10(5), 051403 (2005). 4. J. Requejo-Isidro, J. McGinty, I. Munro, D. S. Elson, N. P. Galletly, M. J. Lever, M. A. A. Neil, G. W. H. Stamp, P. M. W. French, P. A. Kellett, J. D. Hares, and A. K. L. Dymoke-Bradshaw, “High-speed wide-field time-gated endoscopic fluorescence-lifetime imaging,” Opt. Lett. 29(19), 2249–2251 (2004). 5. G. T. Kennedy, H. B. Manning, D. S. Elson, M. A. A. Neil, G. W. Stamp, B. Viellerobe, F. Lacombe, C. Dunsby, and P. M. W. French, “A fluorescence lifetime imaging scanning confocal endomicroscope,” J. Biophotonics 3(1-2), 103–107 (2010). 6. G. O. Fruhwirth, S. Ameer-Beg, R. Cook, T. Watson, T. Ng, and F. Festy, “Fluorescence lifetime endoscopy using TCSPC for the measurement of FRET in live cells,” Opt. Express 18(11), 11148–11158 (2010). 7. G. T. Kennedy, S. Coda, A. J. Thompson, D. S. Elson, M. A. A. Neil, G. W. Stamp, A. Thillainayagam, B. Viellerobe, F. Lacombe, C. Dunsby, and P. M. W. French, “Fluorescence lifetime imaging endoscopy,” Proc. SPIE 7893, 789308 (2011). 8. D. S. Elson, J. A. Jo, and L. Marcu, “Miniaturized side-viewing imaging probe for fluorescence lifetime imaging (FLIM): validation with fluorescence dyes, tissue structural proteins and tissue specimens,” New J. Phys. 9(5), 127 (2007). 9. S. Cheng, J. J. Rico-Jimenez, J. Jabbour, B. Malik, K. C. Maitland, J. Wright, Y. S. Cheng, and J. A. Jo, “Flexible endoscope for continuous in vivo multispectral fluorescence lifetime imaging,” Opt. Lett. 38(9), 1515– 1517 (2013). 10. H. Xie, J. Bec, J. Liu, Y. Sun, M. Lam, D. R. Yankelevich, and L. Marcu, “Multispectral scanning time-resolved fluorescence spectroscopy (TRFS) technique for intravascular diagnosis,” Biomed. Opt. Express 3(7), 1521– 1533 (2012). 11. D. G. Ouzounov, D. R. Rivera, W. W. Webb, J. Bentley, and C. Xu, “Miniature varifocal objective lens for endomicroscopy,” Opt. Lett. 38(16), 3103–3106 (2013). #204162 $15.00 USD Received 10 Jan 2014; revised 19 Feb 2014; accepted 19 Feb 2014; published 26 Feb 2014 (C) 2014 OSA1 March 2014 | Vol. 5, No. 3 | DOI:10.1364/BOE.5.000921 | BIOMEDICAL OPTICS EXPRESS 921 12. S. Shrestha, B. E. Applegate, J. Park, X. Xiao, P. Pande, and J. A. Jo, “High-speed multispectral fluorescence lifetime imaging implementation for in vivo applications,” Opt. Lett. 35(15), 2558–2560 (2010). 13. P. Pande and J. A. Jo, “Automated analysis of fluorescence lifetime imaging microscopy (FLIM) data based on the Laguerre deconvolution method,” IEEE Trans. Biomed. Eng. 58(1), 172–181 (2011). 14. Y. Sun, H. Xie, J. Liu, M. Lam, A. J. Chaudhari, F. Zhou, J. Bec, D. R. Yankelevich, A. Dobbie, S. L. Tinling, R. F. Gandour-Edwards, W. L. Monsky, D. G. Farwell, and L. Marcu, “In vivo validation of a bimodal technique combining time-resolved fluorescence spectroscopy and ultrasonic backscatter microscopy for diagnosis of oral carcinoma,” J. Biomed. Opt. 17(11), 116003 (2012). 15. American National Standards Institute, Safe Use of Lasers, ANSI Z136.1–2007, (Laser Institute of America, 2007). 16. M. Yuvaraj, K. Udayakumar, V. Jayanth, A. Prakasa Rao, G. Bharanidharan, D. Koteeswaran, B. D. Munusamy, C. Murali Krishna, and S. Ganesan, “Fluorescence spectroscopic characterization of salivary metabolites of oral cancer patients,” J. Photochem. Photobiol. B 130C, 153–160 (2013). 17. H. M. Chen, C. P. Chiang, C. You, T. C. Hsiao, and C. Y. Wang, “Time-resolved autofluorescence spectroscopy for classifying normal and premalignant oral tissues,” Lasers Surg. Med. 37(1), 37–45 (2005). 18. Z. Nie, R. An, J. E. Hayward, T. J. Farrell, and Q. Fang, “Hyperspectral fluorescence lifetime imaging for optical biopsy,” J. Biomed. Opt. 18(9), 096001 (2013). 19. J. Bec, D. M. Ma, D. R. Yankelevich, J. Liu, W. T. Ferrier, J. Southard, and L. Marcu, “Multispectral fluorescence lifetime imaging system for intravascular diagnostics with ultrasound guidance: in vivo validation in swine arteries,” J. Biophotonics doi: 10.1002/jbio.201200220 (2013). 20. Y. Sun, J. E. Phipps, J. Meier, N. Hatami, B. Poirier, D. S. Elson, D. G. Farwell, and L. Marcu, “Endoscopic Fluorescence Lifetime Imaging for In Vivo Intraoperative Diagnosis of Oral Carcinoma,” Microsc. Microanal. 19(4), 791–798 (2013).


Introduction
In fluorescence lifetime imaging (FLIM), the fluorescence lifetime is measured at each spatially resolvable location within a fluorescence image [1].FLIM offers a noninvasive approach for characterizing the biochemical composition of biological tissue.In recent years, there has been an increasing interest in evaluating the application of endogenous FLIM for clinical diagnosis.A central issue for the clinical translation of FLIM is the development of compact and fast FLIM endoscopy systems suitable for in vivo tissue imaging.However, the FLIM endoscope designs proposed thus far, including those summarized below, are still not well suited for clinical applications.
The fluorescence lifetime can be measured in the frequency domain by estimating the phase shift or amplitude modulation in a fluorescence signal obtained by modulated excitation of the sample [1].Alternatively, the fluorescence lifetime can be implemented in the time domain by directly measuring the fluorescence decay following pulsed light excitation [1].Wide-field time domain FLIM can be implemented using time-gated imaging.Point scanning time domain FLIM can be implemented using either time-correlated single-photon counting (TCSPC) or the direct pulse recording approach.For the latest, the fluorescence intensity decay is measured as function of time upon a single excitation pulse using high-bandwidth detectors and digitizers.
Several FLIM endoscope designs have already been proposed.Two wide-field FLIM rigid endoscope designs were reported, one in the frequency-domain [2], and the other in the timedomain, both allowing video-rate FLIM imaging [3,4].Two flexible endoscope designs have also been proposed using a scanning FLIM implementation based on TCSPC [5,6].A rigid hand-held FLIM system based on a wide-field time-gating FLIM implementation was recently reported [7].A common limitation of all these designs, however, was the acquisition of the fluorescence emission at a single emission band.
For multispectral FLIM endoscopy, two flexible endoscope designs, based on a wide-field time-gated FLIM implementation have been recently proposed.The first wide-field design, however, presented two main limitation [8].First, since multispectral imaging was sequential, the time required to record FLIM images at multiple emission bands increases with the number of spectral bands.Second, since illumination light is delivered through a fiber adjacent to the imaging bundle, achieving uniform illumination and endoscope compactness can be challenging.The second design addressed these two limitations and was used to image in vivo malignant lesions in a hamster cheek pouch model of oral epithelial cancer [9].However, in order to achieve fast FLIM imaging, a rapid lifetime determination (RLD) algorithm had to be adopted, limiting its temporal resolution.Most recently, a scanning based multispectral FLIM catheter design was presented; however, the level of excitation power required exceeded the maximum permissible exposure (MPE), making it not suitable yet for in vivo human imaging [10].
To overcome some of the limitations of the previous designs, we propose a novel multispectral FLIM handheld rigid endoscope based on the point scanning direct pulse recording implementation, using safe and permissible exposure levels at pixel rate of several tens of kHz.Taking advantage of this relatively high pixel rate, we were able to perform the fastest in vivo multispectral FLIM imaging in the human oral cavity reported thus far.We also report the first demonstration of real-time deconvolution of the instrument response from the fluorescence decay at each pixel of the image, which allowed accurate real-time lifetime map estimation and visualization at multiple spectral bands simultaneously.

System configuration
The system consists of a handheld box (volume: 7 × 13 × 5 cm 3 , mass: 450 g) fitted with a custom-designed rigid endoscope (length: 14 cm, diameter: 1.7 cm) as shown in Fig. 1(a).The schematic of the proposed system is shown in Fig. 1(b).A frequency-tripled Q-switched Nd:YAG laser (355 nm, 1 ns pulse width, 100 kHz max.rep.rate, Advanced Optical Technology) is used as the excitation source.A multimode fiber with core diameter of 25µm (0.10 NA, HPSC25, Thorlabs) or 50 µm (0.22 NA, FVP050055065, Polymicro Technologies) delivers the excitation light to the handheld box.Inside the handheld box, the excitation light is collimated (L2: f = 11 mm, ARC 350-700 nm, CFC-11X-A, Thorlabs) and scanned by a pair of galvonometer mirrors (5 mm beam aperture, ± 5 mechanical degrees, Cambridge Technology) on the proximal end of the rigid endoscope.The rigid endoscope was built using standard 0.5 inch lens tubes and consists of three lenses.The first two lenses from the proximal end (L3, L4: f = 30 mm, LB003, Thorlabs) serve as an image relay, while the third lens (L4: Near UV achromat doublet; f = 30 mm, 50 mm or 60 mm; Edmund Optics) works as the objective.In this configuration, the excitation fiber core diameter and the focal length of the objective determine the lateral resolution and the field of view (FOV) of the system.Finally, a lens tube with length equal to the objective's focal length is added at the distal end, which allows the sample to be placed in contact with the probe.Such a configuration can potentially reduce motion artifact during image acquisition.An optical design simulation (Zemax) of the rigid endoscope optical design was also performed as follows.The excitation wavelength was set to 355 nm and consisted of a point source (object NA = 0.2), a collimator (f = 11 mm, CFC-11X-A, Thorlabs), a pair of mirror scanners (5 mm diameter) and the threelens combination for the endoscope described above.In this configuration, we were able to simulate the FOV by changing the scan angle of the mirror scanners.By using objective lens with focal lengths of 30mm, 50mm and 60mm, we could vary the FOV.The Zemax simulation estimated the numerical aperture of the endoscope at ~0.028 and the maximum FOV at ~5mm for 30mm focal length objective, ~10 mm for the 50 mm focal length objective and ~12mm for the 60 mm focal length objective, which are close to what it was measured experimentally.We also analyzed an intensity line profile across the features of a 1951 USAF resolution target to quantify the lateral resolution.The intensity profile represents the edgeresponse function and its first-order derivative is the line-spread function (LSF) which is also the cross section of the lateral point spread function (PSF) [11].As an example, for the 25 um excitation fiber and the 50 mm focal length objective, the full width at half maximum (FWHM) of the PSF was calculated to be 44.8 um, which is in close agreement with the corresponding qualitatively value determined by imaging the USAF resolution target as shown in Table 1.Table 1 also provides FOV and lateral resolution values for different combinations of fiber core size and objective's focal length.
The time-resolved fluorescence emission is spectrally divided in three separate emission bands and multiplexed in time following a strategy previously reported [12].Briefly, a set of dichroic mirrors (DM2, DM3) and filters (F1-F3) separate the emission into multiple spectral bands, each one coupled into separate multimode fibers of different lengths that provide an optical delay between each spectral band.Thus, for a single excitation pulse, multiple decays corresponding to different spectral bands can be recorded using a single detector.The spectral bands can be customized based on the targeted fluorophores.We selected the 390 ± 20 nm, 452 ± 22.5 nm, and >500 nm bands to optimally distinguish emission from three tissue endogenous fluorophores: collagen, reduced nicotinamide adenine dinucleotide (NADH), and flavin adenine dinucleotide (FAD), respectively.The multispectral fluorescence signal is detected by a multichannel plate photomultiplier tube (MCP-PMT, 25 ps transient time spread, R3809U-50, Hamamatsu), followed by a preamplifier before being digitized at 6.25 GS/s by a high-speed digitizer (PXIe-5185, National Instruments) resulting in a temporal resolution of 320 ps.

Table 1. Comparison of resolution and FOV with different focal length objectives and excitation fiber core diameters
Excitation fiber core dia. (µm) Objective focal length (mm)

Data processing
The multispectral FLIM data consisted of three fluorescence decays per pixel (one per emission band).The relatively long excitation pulse width (FWHM: ~1 ns) necessitated the temporal deconvolution of the instrument response from the measured fluorescence decay in order to obtain accurate estimation of the fluorescence lifetime.Time deconvolution was performed offline using an optimized Laguerre expansion technique algorithm [13].After deconvolution, nine images were generated to quantify the fluorescence emission of the samples: absolute integrated fluorescence intensity, normalized integrated fluorescence intensity, and average lifetime maps for each of the three emission bands.The respective normalized intensity maps were calculated by I 1 /(I 1 + I 2 + I 3 ), I 2 /(I 1 + I 2 + I 3 ), I 3 /(I 1 + I 2 + I 3 ), where I 1 , I 2 and I 3 are the absolute intensity values at corresponding pixels in the three intensity images.The average lifetime τ ave images were calculated via τ ave = ave τ = th(t)/ h(t)   , where h(t) is the deconvolved temporal decay.
In addition, to demonstrate real-time estimation and visualization of the multispectral FLIM maps, an online deconvolution method was applied, in which the recorded timeresolved decay for each pixel was compared against a lookup table of pulses generated by convolving the instrument response with single exponential decays with time constants ranging from 0.2 to 8 ns (in steps of 0.2 ns).The best match in terms of the minimum normalized means squared error provided a single exponential estimation of the fluorescence lifetime.
All the videos showed in this paper were processed with this online method, whereas the figures were processed offline with the Laguerre deconvolution algorithm.

System validation and tissue imaging
The handheld imaging system was first validated with 1 mM solutions of 1,4-bis (5phenyloxazol-2-yl) benzene-POPOP (in ethanol), NADH and FAD (in phosphate buffered saline).The standard dye solutions were loaded in three quartz capillary tubes and placed side by side under the probe.A video was recorded at ~1.33 frames per second.
To demonstrate its ability for in vivo application in animal models, the handheld system was validated by imaging a normal hamster cheek pouch in vivo.The imaging protocol was approved by the Institutional Animal Care and Use Committee at Texas A&M University.During imaging, the hamster was first anesthetized, and a cheek pouch was pulled out and extended, and the rigid probe was gently placed on the mucosal surface.A video was recorded during the movement of the probe on top of the mucosa tissue at ~1.33 frames per second.
The clinical potential of this system was further demonstrated by imaging a human oral biopsy ex vivo and the ventral surface of a human tongue in vivo.The respective imaging protocols were approved by the Institutional Review Boards at Texas A&M University and Baylor College of Dentistry.For the ex vivo study, the oral tissue biopsy was transported immediately in PBS to the imaging system following excision.A single FLIM image was recorded with the epithelium in contact with the probe.One end of the biopsy was sutured to mark the image orientation to allow comparison with the respective histology section.Finally, the tissue specimen was fixed in 10% formalin and processed for hematoxylin and eosin (H&E) histological analysis.For the in vivo study, the ventral tongue region of a human volunteer was imaged.The probe was inserted into the volunteer's month and gently placed on the target location.Only one frame was collected to minimize exposure to UV radiation.
For all the aforementioned experiments, the following working parameters were used.The laser pulse energy at the sample was set at ~1 µJ/pulse, resulting in adequate signal-to-noise ratio (SNR ≥ 30 dB).Since only one pulse is required per pixel, the pixel rate is equal to the laser repetition rate.The laser repetition rate was set at 30 kHz and the total number of pixels per frame was set at 150 × 150, corresponding to an acquisition speed of ~1.33 Hz.The 50 µm excitation fiber and the 50 mm focal objective lens were used, corresponding to a lateral resolution of ~110 µm.

Capillary tubes with fluorophores
For 355 nm excitation wavelength, the emission peaks for POPOP, NADH and FAD are approximately at 390 nm, 450 nm and 540 nm, respectively [12].The normalized intensity maps (Fig. 2(a)) confirmed strong emission of POPOP at both the 390 nm and 452 nm channels, strongest emission of NADH at the 452 nm channel, and emission of FAD only at the >500 nm channel.The average lifetime maps (Fig. 2(b)) were also in good agreement with the previously published values for the corresponding fluorophores [12].A video of the online processed results for the capillary validation is shown in Media 1.The average lifetime values for each fluorophore estimated using the offline and the online methods are compared in Table 2 (calculated pixel-to-pixel for the areas corresponding to the entire capillary).

In vivo hamster cheek pouch
Results from an imaged region are shown in Fig. 3.The FLIM maps indicate strong fluorescence intensity at the 390 nm and 452 nm channels (Fig. 3(a)-3(b)), and lifetime values between 4 and 6 ns (Fig. 3(c)), reflecting a collagen-dominant autofluorescence expected in normal epithelial tissue.Notice that the vasculature network can be observed in the absolute fluorescence intensity maps, as blood absorption attenuates the fluorescence signal.The normalized intensity and lifetime maps, on the other hand, were insensitive to blood absorption, and indicated spatially uniform spectral and lifetime properties of the autofluorescence emission of normal cheek pouch epithelial tissue.This was expected, since both the layered structure and the relative concentration of endogenous fluorophores (NADH and FAD in the epithelium, and collagen in the underlying stroma) are maintained throughout the normal epithelial tissue.A video of the online processed results from another cheek pouch region is shown in Media 2. The average lifetime from these two normal cheek pouch regions estimated using either of the methods (offline or online) are compared in Table 2 (calculated pixel-to-pixel for the area corresponding to the entire FOV).

Ex vivo human oral biopsy
The absolute intensity, normalized intensity and average lifetime maps are shown in Fig. 4(a), 4(b), and 4(c), respectively.For the intensity maps, strong emission was observed at the 452 nm channel, followed by comparatively lower fluorescence at the 390 nm and >500 nm channels.For the average lifetime maps, the 452 nm channel showed relative shorter lifetime than 390 nm channel attributed to NADH, which shows a short lifetime and peak emission at ~450 nm.While most of the FLIM parameters (intensity and lifetime) did not show significant variation across different regions of the biopsy, a small region (marked as Region 1 in Fig. 4(c)) in the lifetime map for >500 nm channel showed larger value of lifetime (5.19 ± 0.30 ns) in contrast to rest of the biopsy.This area was later diagnosed as superficial invasive squamous cell carcinoma.The corresponding histology image is shown in Fig. 4(d).
For comparison, the histology image of a region which represented the rest area of this biopsy (marked as Region 2 in Fig. 4(c)) with relatively lower lifetime values (4.03 ± 0.19 ns) was also included and shown in Fig. 4(e).This particular region in Fig. 4(e) was diagnosed as dysplasia as most of the biopsy.The increase in lifetime in Region 1 can be attributed to poryphirin, the presence of which is known to increase with progression of precancer [14].

In vivo ventral region of human tongue
The absolute intensity, normalized intensity and average lifetime maps are shown in Fig. 5. Strong emission was observed in 452 nm channel, followed by comparatively lower fluorescence at 390 and >500 nm channels.The average lifetime maps showed similar lifetime of 4.04 ± 0.29 ns, 4.22 ± 0.28 ns and 4.19 ± 0.28 ns for 390 nm, 452 nm and >500 nm channels, respectively.Notice also here that spatial contrast was evident in the absolute fluorescence intensity maps, while little contrast was observed in the normalized intensity and lifetime maps, as expected for normal epithelial tissue.

Discussions
As stated before, pulse energy of 1 µJ at the sample provides adequate SNR per pixel; thus, the pixel rate equates to the laser repetition rate.Since the frame rate is proportional to the pixel rate and the number of pixels per frame, imaging speed can be increased by either increasing the laser repetition rate or decreasing the number of pixels per frame.The number of pixels per frame can be reduced without sacrificing lateral resolution if the FOV is also reduced accordingly; thus, higher frame rates can be achieved at smaller FOV.Increasing the laser repetition rate while maintaining the pulse energy will increase the average power transmitted through the excitation fiber and ultimately deposited on the sample.Thus, the maximum average power that can be practically applied is ultimately limited by the damage threshold of both the excitation fiber and the sample.For all the experiments, the laser repetition rate was limited to 30 kHz in order to avoid rendering permanent damage to the input facet of the excitation fiber (observed at higher repetition rates).To increase the frame rate, the use of other excitation fibers capable of handling higher average power levels (thus higher pixel rate) is currently being explored.
In order to use this system in a clinical setting in vivo, the laser energy levels should not exceed the tissue damage threshold which can be estimated in terms of the maximum permissible exposure (MPE) provided by the American National Standards Institute (ANSI) standards for the safe use of lasers [15].The single pulse limit for a 1 ns pulse at 355 nm is 303 µJ for both eye and skin.Therefore, we are limited by the average irradiance in the limiting aperture (3.5 mm diameter) which was calculated to be 29.8 mJ.For our system, the actual energy deposited is 2.8 mJ, which is an order of magnitude lower than the MPE for both ocular and skin use.We are thus confident that in vivo imaging can be performed safely with the proposed handheld FLIM imaging system.
One significant advantage of the proposed multispectral FLIM endoscope design is that, unlike previous implementations [3,4,7,9], it can achieve relatively high imaging speed without sacrificing temporal resolution.The achieved high temporal resolution allows correcting for the non-ideal instrument response through time deconvolution.In addition, since the entire fluorescence decay is measured directly, the full complexity of its temporal dynamics can be captured and is no longer limited to a single exponential approximation.Although our imaging systems allows measuring time-resolved fluorescence data with high temporal resolution, accurate estimation of fluorescence lifetimes still requires computationally expensive iterative deconvolution methods.Our fully validated Laguerre deconvolution method is significantly faster than standard nonlinear least square deconvolution algorithms, but it is still not suitable for online processing and visualization of FLIM data [13].To take advantage of the relative high pixel rate achieved by our system and to demonstrate real-time FLIM data processing and visualization (as shown in Media 1 and Media 2), the online deconvolution method described above was proposed.An interesting observation was the fact that the online processing and visualization of the multispectral FLIM maps were insensitive to movement (as shown in the videos).In Table 2, the values of the online lifetime estimations are compared against the offline estimation obtained with the Laguerre deconvolution method.The reported fluorescence lifetime values correspond to the average fluorescence lifetime.If the fluorescence decay follows a single exponential dynamic, the average fluorescence lifetime and the single exponential time constant should have the same value.However, when the measured fluorescence decay follows more complex dynamics, as it is usually the case for endogenous fluorescence measured in our experiments, the corresponding average fluorescence lifetime will be underestimated is the data is treated as a single-exponential decay.As expected, the online values were underestimated with respect to the offline values due to the single exponential approximation.Nevertheless, the online FLIM maps provided fluorescence lifetime contrast similar to that of the offline FLIM maps.
While the >500 nm emission band was provided to target and isolate fluorescence from FAD, there can be a significant contribution of fluorescence from porphyrin in precancerous and cancerous areas, and can potentially confound the signal in that band.The presence of porphyrin is usually attributed to changes associated with malignant transformation, and wherein the level of porphyrin is usually elevated in the effected tissue [16].In our study, the effect of porphyrin manifests in Fig. 4(c) where the fluorescence lifetime in the region with invasive carcinoma shows a relative increase due to longer lifetime of porphyrin [17].For our next generation FLIM system, we will consider adding another emission window corresponding to porphyrin (peak emission: ~630 nm).In addition to isolating fluorescence from FAD, a fourth spectral window can offer us another parameter to characterize the progression of precancer.
A number of multispectral FLIM implementations have been reported recently.Nie et al. proposed a scanning FLIM implementation based on the direct pulse recording approach capable of performing time-resolved measurements at multiple emission wavelengths with high spectral resolution by means of an AOTF in the collection path [18].The limitation of this design, however, was the relatively slow pixel rate (~3 Hz compared to 10's kHz by our implementation).In addition, an endoscopic implementation of their system has not been yet reported.Bec et al. reported a flexible endoscope design for intravascular imaging based on a very similar pulse sampling and multispectral acquisition technique as ours [19].However, their reported pixel rate was significantly lower than ours (5 kHz vs. 10's kHz).In addition, the level of excitation power required by their design exceeded the MPE, while the total energy deposited by our design was an order of magnitude lower than the MPE.Sun et al. reported intraoperative in vivo imaging of oral cancer using a flexible FLIM endoscope based on a wide-field gated approach using an ICCD [20].Their acquisition time, however, was much longer that ours (2 minutes per spectral band vs. less than 1 s for all spectral bands by our design) for a smaller FOV (4 mm vs. 6.5-13mm by our design).Another important difference between all these designs and ours is our capability to perform real-time deconvolution of the instrument response from the fluorescence decay at each pixel of the image, which allowed real-time lifetime map estimation and visualization at multiple spectral band simultaneously.
As mentioned above, our target application is to image the oral mucosa in vivo.Our clinical collaborators (YC and JW) have indicated that the length of our current endoscope design (14 cm) is appropriate to image the oral cavity, as we have already demonstrated by imaging the normal oral mucosa of human volunteers.In addition, we are confident that coupling a longer rigid or a flexible endoscope to our FLIM imaging system for other applications will not pose any major technical challenges, since scanning at the proximal end of the endoscope of choice can be achieved.Nevertheless, an endoscope design with even smaller size and less weight would be more suitable for handling and to access most of the oral cavity.Thus, we are currently working on alternative endoscope designs based on MEMS scanning mirrors that would further reduce the endoscope size and weight.

Conclusions
In summary, we report the first demonstration of a time-domain multispectral FLIM endoscope that requires only one excitation pulse per pixel; thus, the pixel rate equates to the laser repetition rate.Taking advantage of this relatively high pixel rate, we were able to perform the fastest in vivo multispectral FLIM imaging in the human oral cavity reported thus far.Finally, we also report the first demonstration of real-time deconvolution of the instrument response from the fluorescence decay at each pixel of the image, which allowed accurate real-time lifetime map estimation and visualization at multiple spectral bands simultaneously.This design will facilitate the evaluation of multispectral FLIM for in vivo applications, and is currently being explored to image oral cancer, which is our target clinical application.

Fig. 3 .
Fig. 3.In vivo validation imaging of a normal hamster cheek pouch: (a) Absolute integrated fluorescence intensity maps, (b) Normalized integrated fluorescence intensity maps, and (c) Fluorescence lifetime maps.Continuous in vivo imaging of the cheek pouch was also demonstrated (Media 2).FOV: 10 × 10 mm 2 .

Fig. 4 .
Fig. 4. Ex vivo human oral biopsy: (a) Absolute integrated fluorescence intensity maps, (b) Normalized integrated fluorescence intensity maps, and (c) Fluorescence lifetime maps.FOV: 10 × 10 mm 2 .The thread at the bottom of (a), (b) and (c) was the suture used to mark the tissue orientation for histology study.(d) Histology image for the position marked in (c) as Region 1 which was diagnosed as superficial invasive squamous cell carcinoma.(e) Histology image for the position marked in (c) as Region 2 which was diagnosed as dysplasia.Both Region 1 and Region 2 are squares of 600um by 600um.The scale bars in both (d) and (e) represent 200 µm.

Fig. 5 .
Fig. 5.In vivo imaging of the ventral tongue from a normal human volunteer: (a) Absolute integrated fluorescence intensity maps, (b) Normalized integrated fluorescence intensity maps, and (c) Fluorescence lifetime maps.FOV: 10 × 10 mm 2 .