Digital micromirror device as a spatial illuminator for fluorescence lifetime and hyperspectral imaging Artur Bednarkiewicz,* Mounir Bouhifd, and Maurice P. Whelan Nanotechnology and Molecular Imaging Unit, Institute for Health and Consumer Protection, European Commission Joint Research Centre, Via E. Fermi 1, 21020 Ispra, Varese, Italy *Corresponding author:
[email protected] Received 6 September 2007; accepted 13 December 2007; posted 31 January 2008 (Doc. ID 87254); published 17 March 2008
Time-domain fluorescence lifetime imaging (FLIM) and hyper-spectral imaging (HSI) are two advanced microscopy techniques widely used in biological studies. Typically both FLIM and HSI are performed with either a whole-field or raster-scanning approach, which often prove to be technically complex and expensive, requiring the user to accept a compromise among precision, speed, and spatial resolution. We propose the use of a digital micromirror device (DMD) as a spatial illuminator for time-domain FLIM and HSI with a laser diode excitation source. The rather unique features of the DMD allow both random and parallel access to regions of interest (ROIs) on the sample, in a very rapid and repeatable fashion. As a consequence both spectral and lifetime images can be acquired with a precision normally associated with single-point systems but with a high degree of flexibility in their spatial construction. In addition, the DMD system offers a very efficient way of implementing a global analysis approach for FLIM, where average fluorescence decay parameters are first acquired for a ROI and then used as initial estimates in determining their spatial distribution within the ROI. Experimental results obtained on phantoms employing fluorescent dyes clearly show how the DMD method supports both spectral and temporal separation for target identification in HSI and FLIM, respectively. © 2008 Optical Society of America OCIS codes: 170.0110, 170.6280, 170.6920, 230.6120.
1. Introduction
Fluorescence imaging is a powerful technique for the analysis of biological samples. In the simplest case, an imaging system, such as an epifluorescence microscope equipped with an excitation lamp and a filter cube, is used to illuminate the sample at one wavelength while imaging it at another. However, much additional information can be obtained from the same sample if the more advanced techniques of hyper-spectral imaging (HSI) and fluorescence lifetime imaging (FLIM) are employed. In a HSI setup the complete fluorescence spectrum is acquired for every point of the object. It is a particularly useful method to clearly discriminate different structural features within a sample when multiple staining with many fluorescent dyes is employed whose spectra overlap 0003-6935/08/091193-07$15.00/0 © 2008 Optical Society of America
[1]. Rather than measuring (spectral) intensity, FLIM involves the mapping of the decay time of fluorescence [2], which can vary depending on the local environment. Such a functional imaging approach is well suited to the study of biochemical and biophysical processes in tissues and cells, often in a nondestructive manner [3]. In this paper we propose a novel experimental setup to carry out HSI or FLIM in a practical and inexpensive fashion, which is based on the use of a digital micromirror device (DMD) as a light modulator for excitation. Wide-field spectral imaging is usually performed using a set of optical emission filters mounted on a filter wheel or rotary filter-cube holder. This approach is straightforward and relatively cheap, but is usually rather slow and cumbersome due to the mechanical switching of the filters. Typically only a limited number of spectral bands can be acquired, which allows only partial separation of overlapping fluorescence spectra. More versatile HSI systems 20 March 2008 / Vol. 47, No. 9 / APPLIED OPTICS
1193
employ electro-optic devices such as liquid crystal tunable filters (LCTFs) [4], linearly variable bandpass dielectric filters (LVDF) [5], or acousto-optical tunable filters (AOTF) [6]. The LCTF solution offers approximately 30% of passband transmission efficiency and spectral resolution of a few nanometers. Spectral resolution of a LVDF device is roughly 15 nm, its spectral range covers 400–700 nm, and its transmission efficiency is around 40%. The AOTF is an electronically controllable, variable bandwidth (down to 1 nm) optical filter that provides significant versatility and performance in comparison with other tunable filters. It supports random access to any transmission-band or continuous spectral tuning and thus is very suitable for HSI when combined with a sensitive camera. However, all these widefield HSI systems based on tunable filters can suffer from relatively poor image quality due to the scattering present in most biological samples. This can result in image contrast decrease and loss of quantitative information, such as the concentration of a fluorophore. As an alternative, reconstruction of a fluorescence map making point-by-point measurements has been proved to deliver images of superior quality [7,8], which reveal more accurately the localization and concentration of fluorescent markers. However, since this approach usually requires raster scanning carried out by intricate electromechanical systems it is difficult and costly to implement. Fluorescence lifetime imaging in the time-domain can generally be approached in two ways, namely, by wide-field fluorescence lifetime imaging [2] or by point-by-point raster scanning. Wide-field techniques employ a gated optical image intensified (GOI) camera in combination with a high-power pulsed laser illuminating the whole field of view. Such a FLIM system is fast [2,9] but demonstrates lower temporal and spatial resolution in comparison with scanning measurements. The high cost and lack of portability are also issues that have limited its broad use. On the other hand, the successful demonstration of GOI-based FLIM using picosecond laser diodes (LDs) has helped matters somewhat [10]. The second approach to FLIM exploits the timecorrelated single-photon counting (TCSPC) technique [2] combined with point-by-point scanning [11] or laser confocal scanning [12]. Picosecond LDs are the most popular excitation source because of their compactness and reasonable cost, while a photomultiplier tube (PMT) or an avalanche photodiode (APD) are used to count the emitted fluorescence photons. With respect to wide-field FLIM, scanning TCSPC is more straightforward and less expensive to implement. It also offers a higher temporal resolution, but at the price of much slower imaging rates. To obtain FLIM images, either the sample is moved on a translation stage under the excitation spot or the sample is held steady while the laser beam is raster scanned by using galvanomirrors. Although a number of attractive commercial solutions exist based on both scanning methods, the disadvantages of 1194
APPLIED OPTICS / Vol. 47, No. 9 / 20 March 2008
point scanning in general are its low imaging rate and the lack of flexibility in how an image is formed. Normally a minimum sampling time of 100 μs is required per measurement point (pixel), thus leading to a duration of several seconds to tens of minutes to acquire a complete image. The actual duration depends on a number of factors including the strength of the fluorescence signal and the desired temporal and spatial resolution. Most scanning systems are analog by nature and thus offer neither random access to any point or region of interest (ROI) on the sample nor the possibility to bin the photons emitted from different locations on line. At the heart of the HSI-FLIM system described here is a DMD that is a 2D matrix of tiltable micromirrors (each 13:68 μm × 13:68 μm in size) that are controlled independently by the underlying complementary metal-oxide semiconductor (CMOS) electronics [13]. Such DMD technology is well known in multimedia projectors and has already been applied to optical sectioning in fluorescence microscopy [14] and for frequency-domain time-resolved studies [15]. To our knowledge, however, this is the first report on a DMD being employed as a spatial excitation light modulator for either HSI or TCSPC-based FLIM. The setup proposed for combined HSI and FLIM is presented together with the results obtained from simple fluorescence phantoms. 2. Experimental Setup for Hyperspectral and Fluorescence Lifetime Imaging
The laboratory setup to perform HSI and FLIM is illustrated in Fig. 1. Expanded and collimated light from a 405 nm pulsed LD (Model LDH-405 with PDL800-B driver from PicoQuant GmbH, Pmax ¼ 1 mW, τFWHM ¼ 50 ps, 40 MHz repetition rate) was projected
Fig. 1. (Color online) Experimental setup. Light from the laser diode (LD) was expanded to illuminate the DMD through the variable gray filter (VGF). The pattern generated by the DMD was projected onto the sample, and whole-field fluorescence was collected by either a spectrophotometer (HSI experiments) or a photomultiplier PMT (for further TCSPC photon counting dedicated for FLIM experiments).
onto the surface of the DMD spatial light modulator. A DMD kit from Texas Instruments was employed that consisted of a 0.7 XGA chip (Model 714, 1024 × 768 mirror pixels, 13:68 μm=mirror, UV reflection coating) mounted on a Discovery TM-1100 controller board. Binary image patterns were uploaded to the device through a fast USB 2.0 port using a customized software driver (Tyrex Services Group Ltd.) written for LabVIEW (National Instruments). The light reflected by the DMD mirrors was projected through a telecentric system (magnification 1) with spatial filter (adjustable aperture) and through a filter cube (Olympus U-MWBV2) onto the sample, thus exciting the fluorescence. The fluorescence signal emitted by the sample was reflected by a dichroic mirror, passed through a high-pass emission filter and was focused onto the detector. For spectral measurements, the fluorescence light was coupled into a 1 mm diameter multimode optical fiber and detected by a miniature CCD spectrofluorometer (Model SD2000 from Ocean Optics). For time-resolved studies, a photomultiplier tube (PMT, Model R6060-02 from Hamamatsu) powered by a high-voltage (HV) module (Model PS325 from Stanford Research Systems) was used as the detector. The signal from the PMT was amplified and inverted by a high-speed preamplifier module (PAM, Model 102-T from PicoQuant GmbH) and then acquired by a single-photon counting board (Model TimeHarp 200 from Picoquant GmbH) housed in the PC. Switching between each detection system for either HSI or FLIM was done manually. ActiveX components to control the DMD board and acquire data from the spectrophotometer or the photon counting board were used to build a PC software application (LabVIEW version 7.0, National Instruments). The DMD controller offers random and parallel access to any DMD mirror, but only sweeping the sample with excitation beam by sequentially switching ON and OFF respective micromirrors or groups of micromirrors was used here. In its ON state, a micromirror was tilted in order to reflect the excitation light through the projection optics and filter cube onto the sample. The excitation light reflected from the micromirror in the OFF state was dumped and did not hit the sample. While the sample was illuminated with an excitation beam spot, the spectrophotometer or the PMT-TCSPC detection was used to acquire emitted fluorescence and to generate a full HSI or FLIM images, respectively. The maximum spatial resolution (1024 × 768 pixels) was achieved when each individual mirror was sequentially switched on to illuminate a single spot on the object. To boost system sensitivity and to improve signal-to-noise ratio, binning was also possible by addressing a number of neighboring mirrors together. The processing of the acquired signals and the experiments carried out to demonstrate the performance of the system are described in Section 3.
3.
Experimental Investigation
A.
Fluorescence Hyperspectral Imaging
For a demonstration of DMD-based HSI a simple phantom was prepared [Fig. 2(a)] consisting of green (G) and orange (O) fluorescence dyes (obtained from highlighter pens) deposited as spots on white paper (P). The system was configured for raster-scanning mode and for each illumination bin (8 × 8 mirrors) a full spectrum was recorded by the spectrophotometer with a 1 s=bin integration time. For spectral separation of the dyes and subsequent visualization of spectrally similar regions, three distinct wavelength bands were chosen based on the spectra of the dyes measured beforehand [Fig. 2(b)]. The red band (λR ¼ 582–592 nm) was centered on the emission peak of the orange dye, whereas the blue (λB ¼ 477–480 nm) and green (λG ¼ 531– 541 nm) bands were chosen to possibly reduce the cross talk of monitored fluorescence intensities between the green dye and the paper (broad band observed in all spectra at 490 30 nm). Additionally a
Fig. 2. (Color online) (a) Image of the HSI phantom consisting of spots of green (G) and orange (O) dyes on a white-paper background (P). (b) Fluorescence spectra of the dyes and paper used for the phantom and the RGB spectral bands used for spectral separation and visualization. 20 March 2008 / Vol. 47, No. 9 / APPLIED OPTICS
1195
background noise signal averaged over the 708–718 nm band was subtracted from all the spectra before processing. The integral intensities (I) for the three spectral bands (λR , λG , and λB ) were then scaled to fit an 8-bit intensity range, preserving the relative pixel-to-pixel intensity ratio. To visualize the spectral map the three channels of an red–green– blue (RGB) color image were filled according to the scheme R ¼ IðλR Þ, G ¼ IðλB Þ, and B ¼ IðλG Þ. The results obtained for the DMD-based HSI of the phantom are shown in Fig. 3. The RGB color image, which was constructed by using the three-band spectral separation algorithm, is shown in Fig. 3(a) in its raw form, without any postprocessing. One can notice good separation of the orange fluorescent spot from the background, while there is some spectral overlap between fluorescence of the green dye and the paper, manifested in a lower image contrast. B. Fluorescence Lifetime Imaging
For demonstration of time-domain fluorescence lifetime imaging a second simple phantom was prepared [Fig. 4(a)], which consisted of spots of green (G) and orange (O) fluorescent dyes on a background of white paper, plus an additional nonfluorescent black spot (BL). A scanning mode was employed with a bin size of 16 × 16 DMD mirrors, with the acquisition time to record a lifetime curve equal to 1:2 s=bin. The synchronization count rate was equal to 2 × 107 cts=s, while the total count rate was kept at roughly 9 × 104 cts=s to avoid PMT saturation. The HV supply of the PMT was kept constant at 710 V
Fig. 3. (Color online) On-line processed HSI image of the phantom (a) and the image divided into three components according to indications given in Fig. 2(b). Images (b), (d), and (c) show R, G, and B planes corresponding to orange dye, green dye, and paper fluorescence, respectively. 1196
APPLIED OPTICS / Vol. 47, No. 9 / 20 March 2008
Fig. 4. (Color online) (a) White-light image of a phantom (14 mm × 12 mm in size) for FLIM measurements and (b) fluorescence decay curves acquired a priori for orange (1:303 ns) and green (0:812 ns) fluorescence dyes as well as for the paper (0:503 ns).
for all measurements and when necessary the amount of excitation light was adjusted using a variable neutral-density filter to avoid saturation. Similar experiments were also successfully carried out using alternative acquisition times of 0.2 and 0:02 s=bin to examine the trade-off between scan speed and image quality. The fluorescence decay curves acquired for each location were analyzed by two different methods, namely by a rapid lifetime assessment (RLA) algorithm and by a single exponential fit using a Levenberg–Marquard algorithm (LMA). For RLA, four discrete time points were first defined based on decay curves individually acquired for the dyes and the paper prior to imaging [Fig. 4(b)]. Time point tBG corresponded to the region before the rising slope and the average counts DBG around that point (in the 〈tBG − 0:2 ns; tBG þ 0:2 ns〉 time window) was used as a measure of the background photon counts. The three other time points were fixed to equally divide the region between the approximate start (t0 ¼ 4 ns) and end (t2 ¼ 8 ns) of the fluorescence single exponential decay curve recorded for the orange dye. More detailed discussion about the accu-
racy of rapid fluorescence lifetime estimation techniques can be found elsewhere [16]. The averaged fluorescence lifetime 〈t〉 for each measurement was calculated by using the simple formula [17] 〈t〉 ¼ −Δt= lnðD1 =D0 Þ;
ð1Þ
where Δt ¼ t2 − t1 ¼ t1 − t0 (here Δt ¼ 2 ns) is the time window width and D1 and D2 are the integrated photon counts measured for the two time windows, respectively, after subtraction of the background signal DBG . All parameters (i.e., fluorescence intensity, background intensity, and fluorescence lifetime) were stored as separate color planes of a RGB color image. The red channel was used to represent fluorescence intensity corresponding to the total number of photon counts (D1 plus D2 ) recorded for a bin, normalized with respect to the maximum photon counts and mapped onto an 8-bit scale. The green channel encoded the background signal, while the blue channel contained the calculated decay constants. No further image processing or filtering was performed on the constructed RGB image. The RLA processing was fast enough to be carried out on line, and thus the progressive build up of the RBG-encoded lifetime image could be visualized during data acquisition. The Levenberg–Marquard fitting algorithm employs a least-squares approach to determine a set of coefficients (P) to reliably model the experimental dataset with a nonlinear function yðt; PÞ ¼ Pð0Þ expð−t=Pð1ÞÞ þ Pð2Þ. The LMA was implemented here using a customized LabVIEW Non-Linear Lev-Mar Fit.vi virtual instrument. No instrument response function (IRF) was taken into account in these calculations. A detailed analysis of the accuracy of LMA in the low photon count regime can be found elsewhere [18]. The visualization of the LMA results was performed as for the RLA analysis described above. The fluorescence decay image for the RLA algorithm is presented on Fig. 5(a). Although different dyes could be clearly distinguished based on their different fluorescence lifetimes, the RLA gives rise to some artifacts. As can be found from the histogram of lifetimes [Fig. 5(b)], the method indicates the presence of four lifetime components rather than the three expected (i.e., green and orange dyes plus paper). The averaged fluorescence lifetime τi and full width at half-maximum ωi of the bands were measured to be τi ¼ 0:418, 0.477, 0.694, 1:146 ns and ωi ¼ 0:213, 0.057, 0.158, 0.273, respectively. The FLIM results obtained by using the singleexponential LMA fitting approach are presented in Fig. 6. Although it took longer to produce the FLIM image with respect to RLA, the image [Fig. 6(a)] demonstrated better quality, and the histogram of recorded lifetimes [Fig 6(b)] clearly showed three distinct lifetime components as expected. The averaged fluorescence lifetime τi and full width at halfmaximum ωi of the bands were measured to be
Fig. 5. (Color online) (a) False colored FLIM image constructed using the RLA method and RGB encoding; (b) histogram of associated fluorescence decay times.
τi ¼ 0:499, 0.673, 1:174 ns and ωi ¼ 0:078, 0.222, 0.281, respectively. 4.
Discussion
The HSI images presented in Fig. 3 demonstrate how even a basic three-band spectral separation algorithm can distinguish between different fluorophores present in the sample. More advanced approaches, such as separation by spectral fitting to determine the relative contribution of each dye, could be applied for a higher degree of partitioning. Although this would require more computing effort, it would likely reduce the cross talk between the paper and green dye [Fig. 3(d)] as observed here, and provide the basis for more precise quantitative measurements. The FLIM images presented in Figs. 5(a) and 6(a) and their corresponding lifetime distributions [Figs. 5(b) and 6(b), respectively] demonstrate that the fluorescence lifetimes of the green and orange dyes and of the paper can be easily distinguished. One can, however, notice that the fluorescence lifetime measured for the orange dye differs slightly from spot to spot. Therefore relatively broad fluorescence decay distributions (ω4 ∼ 0:27) of the longest component coming from fluorescence of orange dye can be observed. We can explain the observation 20 March 2008 / Vol. 47, No. 9 / APPLIED OPTICS
1197
Fig. 6. (Color online) (a) False colored FLIM image constructed using the LMA method and RGB encoding; (b) histogram of associated fluorescence decay times.
as being the result of nonhomogenous illumination of the sample. It is well known that least-squares-based methods underestimate fluorescence lifetimes for smaller photon counts [18]. Rapid lifetime estimation (RLE) algorithms, based on decay curve integration are also fluorescence intensity dependent. The problems can be avoided or at least minimized by improving the degree of homogeneity of the light illuminating the DMD chip, ideally aiming for a top-hat intensity profile. Alternatively, appropriate dithering of DMD mirrors could be used to vary the exposure time at any spatial location in order to compensate for nonhomogeneous illumination. Such an approach, termed controlled light exposure microscopy (CLEM) was originally designed to reduce photobleaching [19]. The images in Figs. 5(a) and 6(a) also show a yellow-green fringe around the periphery of the orange spots, which also contributes to the broadening of the distribution of the fluorescence decay constant. This feature arises from binning several mirrors together and averaging the signals coming from both the short-living fluorescence of the paper and the long-living fluorescence of the orange dye. Both signal analysis methods used for calculating decay constants only considered the decaying part of 1198
APPLIED OPTICS / Vol. 47, No. 9 / 20 March 2008
the experimental dataset, and thus no IRF was taken into account. As a consequence, the selection of the time window and the amount of data points had an impact on the fluorescence decay parameters obtained with both analysis algorithms. Additionally, it was observed that the actual choice of initial fitting parameters (t0…3 , tBG , Δt) supplied for the first iteration of the LMA algorithm had a slight effect on the calculated lifetimes and their distributions. Nevertheless, the LMA generated fewer artifacts in comparison with the RLA approach. In particular, it is noticeable when comparing Figs. 5(b) and 6(b) that the first broad peak centered at e0:4 ns identified by the RLA method was not present when the more accurate LMA algorithm was applied. On the other hand, while the analysis of single decay based on RLA took roughly 4 ms per bin, the same analysis for LMA took more than 200–600 ms=bin and depended on the initial fitting parameters and signalto-noise ratio of the experimental decay. DMD-FLIM and DMD-HSI require illumination of the whole DMD chip, similar to FLIM or HSI implemented using gated intensified cameras or tunable filters. This is often a challenge when excitation light intensity is limited or sophisticated technical requirements for the excitation source are required, as for picosecond pulses with a megahertz repetition rate. As demonstrated here, a compact pulsed LD source can be successfully applied to reconstruct images of fluorescence decays or fluorescence spectra in macrophantoms larger than 1 cm2 in size. Nevertheless, application of LD (or matrices of semiconductor light sources) with characteristic poor beam quality and high astigmatism needs a lot of attention to correctly expand and homogenize the excitation beam. The communication interface (USB) and software drivers used for this study supported a maximum scanning rate of 48 pixels/s. Due to light losses, the acquisition time was set to around 1 s=bin. However, with careful optimization of the optical path it is believed that a 1 ms=bin acquisition time or less could be obtained. The total time required to capture a full resolution image (0.7 XGA, 1024 × 768 ¼ 786; 432 mirrors or pixels) can then be calculated as 16,384 s (roughly 274 min), which is obviously long. The scanning speed could be significantly improved if a high-speed DMD controller board was used to control pattern projection [20]. Such a solution would offer over 8000 bins=s scanning rate in binary mode. Another approach to increase frame rate is to increase the number of mirrors per bin. Although this results in reduced spatial resolution, it has the benefit that for intensity dependent measurements larger bins increase both excitation and fluorescence intensities, improve the signal-to-noise ratio, and as a consequence allow the shortening of signal acquisition times. Another way to improve accuracy and decrease analysis time is to apply a global analysis (GA) [21] approach to supply the first iteration of the fitting algorithms with good initial para-
meters. The implementation of GA using the DMDFLIM platform is described elsewhere [22]. 5. Conclusions
We have demonstrated a novel implementation of hyper-spectral imaging (HSI) and fluorescence lifetime imaging (FLIM) based on a DMD spatial illumination-light modulator. A miniature spectrophotometer and a single-channel time correlated single photon counting board were used to collect HSI and FLIM images respectively. A picosecond LD was used as an excitation light source, which was sufficient to image simple phantoms as large as 14 mm × 12 mm. The advantages of using a DMD as a spatial illuminator include random, parallel, and rapid access to single pixels or groups of pixels at incongruent locations on the object surface. Pixel or mirror binning can be easily implemented on the DMD to achieve an optimum trade-off between spatial resolution, signal-to-noise ratio, and scanning speed. Such features, when combined with on-line data analysis, provide an ideal platform for the implementation of a GA approach to FLIM. This should result in a significant increase in the speed of curvefitting in comparison with common raster-scanning methods, while preserving the highest spectral and temporal resolution achievable. Overall the DMD illuminator represents a versatile and flexible illumination system for many microscopic and macroscopic HSI- and FLIM-based bioimaging applications, providing an attractive balance between simplicity, performance, and cost. References 1. T. Zimmermann, “Spectral imaging and linear unmixing in light microscopy,” Adv. Biochem. Eng./Biotechnol. 95, 245– 265 (2005). 2. J. R. Lakowicz, Principles of Fluorescence Spectroscopy (Kluwer Academic, 1999). 3. D. Elson, J. Requejo-Isidro, I. Munro, F. Reavell, J. Siegel, K. Suhling, P. Tadrous, R. Benninger, P. Lanigan, J. McGinty, C. Talbot, B. Treanor, S. Webb, A. Sandison, A. Wallace, D. Davis, J. Lever, M. Neil, D. Phillips, G. Stamp, and P. French, “Time domain fluorescence lifetime imaging applied to biological tissue,” Photochem. Photobiol. Sci. 3, 795–801 (2004). 4. N. Gat, “Imaging spectroscopy using tunable filters: a review,” Proc. SPIE 4056, 50–64 (2000). 5. P. R. Barber, B. Vojnovic, G. Atkin, F. M. Daley, S. A. Everett, G. D. Wilson, and J. D. Gilbey, “Applications of cost-effective spectral imaging microscopy in cancer research,” J. Phys. D 36, 1729–1738 (2003). 6. M. Bouhifd, M. P. Whelan, and M. Aprahamian, “Fluorescence imaging spectroscopy utilising acousto-optic tunable filters,” Proc. SPIE 5826, 185–193 (2005). 7. B. W. Pogue, S. L. Gibbs, B. Chen, and M. Savellano, “Fluorescence imaging in vivo: raster scanned point source imaging provides more accurate quantification than broad beam geometries,” Technol. Cancer Res. Treat. 3, 15–21 (2004).
8. N. Ramanujam, J. X. Chen, K. Gossage, R. Richards-Kortum, and B. Chance, “Fast and noninvasive fluorescence imaging of biological tissues in vivo using a flying spot scanner,” IEEE Trans. Biomed. Eng. 48, 1034–1041 (2001). 9. 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. DymokeBradshaw, “High-speed wide-field time-gated endoscopic fluorescence-lifetime imaging,” Opt. Lett. 29, 2249–2251 (2004). 10. D. S. Elson, J. Siegel, S. E. D. Webb, S. Lévêque-Fort, M. J. Lever, P. M. W. French, K. Lauritsen, M. Wahl, and R. Erdmann, “Fluorescence lifetime system for microscopy and multiwell plate imaging with blue picosecond diode laser,” Opt. Lett. 27, 1409–1411 (2002). 11. Y. Zhang, S. A. Soper, L. R. Middendorf, J. A. Wurm, R. Erdmann, and M. Wahl, “Simple near-infrared time-correlated single photon counting instrument with a pulsed diode laser and avalanche photodiode for time-resolved measurements in scanning applications,” Appl. Spectrosc. 53, 497–504 (1999). 12. M. Kress, T. Meier, R. Steiner, F. Dolp, R. Erdmann, U. Ortmann, and A. Rück, “Time-resolved microspectrofluorometry and fluorescence lifetime imaging of photosensitizers using picosecond pulsed diode lasers in laser scanning microscopes,” J Biomed. Opt. 8, 26–32 (2003). 13. D. Dudley, W. M. Duncan, and J. Slaughter, “Emerging digital micromirror device (DMD) applications,” Proc. SPIE 4985, 14– 25 (2003). 14. Q. S. Hanley, O. J. Verveer, and T. M. Jovin, “Optical sectioning fluorescence spectroscopy in a programmable array microscope,” Appl. Spectrosc. 52, 783–789 (1998). 15. Q. S. Hanley, K. A. Lidke, R. Heintzmann, D. J. Arngt-Jovin, and T. M. Jovin, “Fluorescence lifetime imaging in an optically sectioning programmable array microscope (PAM),” Cytometry 67A, 112–118 (2005). 16. K. K. Sharman and A. Periasamy, “Error analysis of the rapid lifetime determination method for double-exponential decays and new windowing schemes,” Anal. Chem. 71, 947–952 (1999). 17. S. P. Chan, Z. J. Fuller, J. N. Demas, and B. A. De Graff, “Optimized gating scheme for rapid lifetime determinations of single-exponential luminescence lifetimes,” Anal. Chem. 73, 4486–4490 (2001). 18. G. Nishimura and M. Tamura, “Artefacts in the analysis of temporal response functions measured by photon counting,” Phys. Med. Biol. 50, 1327–1342 (2005). 19. R. A. Hoebe, C. H. Van Oven, T. W. J. Gadella, P. B. Dhonukshe, C. J. F. Van Noorden, and E. M. M. Manders, “Controlled lightexposure microscopy reduces photobleaching and phototoxicity in fluorescence live-cell imaging,” Nat. Biotechnol. 25, 249–253 (2007). 20. R. Höfling and E. Ahl, “ALP: universal DMD controller for metrology and testing,” Proc. SPIE 5289B, 322–329 (2004). 21. S. Pelet, M. J. R. Previte, L. H. Laiho, and P. T. C. So, “A fast global algorithm for fluorescence lifetime imaging microscopy based on image segmentation,” Biophys. J. 87, 2807–2817 (2004). 22. A. Bednarkiewicz and M. P. Whelan, “Global analysis of microscopic fluorescence lifetime images using spectral segmentation and a digital micromirror spatial illuminator,” J. Biomed. Opt. (to be published).
20 March 2008 / Vol. 47, No. 9 / APPLIED OPTICS
1199
