NIR Fluorescence lifetime macroscopic imaging with a time-gated SPAD camera

Abstract

The performance of SwissSPAD2 (SS2), a large scale, widefield time-gated CMOS SPAD imager developed for fluorescence lifetime imaging, has recently been described in the context of visible range and fluorescence lifetime imaging microscopy (FLIM) of dyes with lifetimes in the 2.5 – 4 ns range. Here, we explore its capabilities in the NIR regime relevant for small animal imaging, where its sensitivity is lower and typical NIR fluorescent dye lifetimes are much shorter (1 ns or less). We carry out this study in a simple macroscopic imaging setup based on a compact NIR picosecond pulsed laser, an engineered diffuser-based illumination optics, and NIR optimized imaging lens suitable for well-plate or small animal imaging. Because laser repetition rates can vary between models, but the synchronization signal frequency accepted by SS2 is fixed to 20 MHz, we first checked that a simple frequency-division scheme enables data recording for different laser repetition rates. Next, we acquired data using different time gate widths, including gates with duration longer than the laser period, and analyzed the resulting data using both standard nonlinear least-square fit (NLSF) and phasor analysis. We show that the fixed synchronization rate and large gate widths characterizing SS2 (10 ns and over) are not an obstacle to accurately extracting lifetime in the 1 ns range and to distinguishing between close lifetimes. In summary, SS2 and similar very large gated SPAD imagers appear as a versatile alternative to other widefield time-resolved detectors for NIR fluorescence lifetime imaging, including preclinical molecular applications.

1. INTRODUCTION

Fluorescence lifetime imaging provides a wealth of information on the molecular environment of a dye and is increasingly used for biological microscopic investigations, thanks to the increased availability of confocal microscopes equipped with pulsed lasers and time-correlated single-photon counting (TCSPC) detectors and electronics. While this approach to fluorescence lifetime imaging microscopy (FLIM) is popular, it is by no means the only one. In particular, widefield detection geometries using intensified CCD (ICCD) cameras have been, and continue to be used, either using frequency-modulated excitation sources and corresponding demodulation schemes, or pulsed excitation sources and time-gated detection. The latter approach yields data that is very similar to that obtained with standard TCSPC approaches, thanks to the very short time gate achievable with modem electronics, with the caveat that it is much less photon-efficient than single-photon counting detector-based approaches.

The advent of CMOS single-photon avalanche diode (SPAD) arrays (SPAD cameras) holds the promise of combining the best of both worlds, TCSPC measurement and widefield imaging, enabling faster and less damaging observation of samples, with obvious advantages. However, SPAD cameras are at this point still limited in their number of pixels and are transferring histogrammed decay data, due to data bandwidth constraints. An alternative design enabling much larger SPAD cameras relies on time-gating instead of time-stamping, due to the lower complexity of the electronics circuit needed to implement single-SPAD gating.

In previous works, we have explored the visible range FLIM capabilities of SwissSPAD2 (SS2), one such SPAD camera comprised of 512 × 512 time-gated SPADs (16.8 μm pitch), characterized by relatively long gate duration (from ~10 ns to ~22 ns) [1, 2]. Using phasor analysis to process the acquired data, we showed that the detector’s performance was limited by shot noise and enabled accurate lifetime measurements at close to video rate [1]. These measurements were performed with a laser repetition rate of 20 MHz, corresponding to the maximum frequency supported for the triggering signal (or SYNC signal) of SS2.

To extend the breadth of applications of SS2 to in vivo imaging, a number of challenges needed to be overcome. Firstly, the detection efficiency of SS2’s SPADs is decreasing with increasing wavelength, dropping from 40% at 600 nm, to 26% at 700 nm and 13% at 800 nm. The detector’s performance being shot noise-limited, this means that longer integration times will be required to achieve sufficient signal-to-noise ratio (SNR). This of course means that dark count noise will increase as well, decreasing the signal-to-background ratio (SBR), with detrimental effect on the measurement precision. These effects are also compounded by the fact that macroscopic imaging uses optics with lower collection efficiency than does microscopic imaging with high numerical aperture objective lenses. Secondly, NIR dyes are characterized by significantly shorter lifetimes than visible dyes, with lifetimes of the order of 1 ns being typical. This gets close to the response time of the SPAD (measured by the rise and fall times of the gate profile, of the order of 400 ps and 600 ps, respectively) and could require higher SNR than for visible dyes with longer lifetimes, in order to obtain reliable measurements. Finally, high power pulsed lasers typically used for NIR in vivo FLI, such as Ti:Sa lasers, are generally characterized by high repetition rates (80 MHz or higher), requiring some kind of workaround to be compatible with the SYNC rate limitations of SS2.

Here, we report our preliminary characterization of SS2 for lifetime measurements in the NIR range, using a custom-designed macroscopic imaging setup described in the next section. We benchmark the performance of this setup using solution of biomedically relevant NIR dyes (ICG and IRDye800CW) and report our macroscopic FLI (MFLI) results in different gate and laser repetition rate configurations, using phasor analysis and standard nonlinear least square fitting (NLSF). Actual applications of SS2 following this characterization have been recently reported [3].

2. SETUP AND EXPERIMENT DESCRIPTION

2.1. Optical Setup

Figure 1A shows a schematic of the setup, while Figure 1B is a photograph of the actual imaging part of the setup, enclosed in a light-tight box. The output of a 765 nm ps-pulsed laser (VIS-IR-765, Picoquant, USA) with adjustable repetition rate (80, 40, 20, etc. MHz) was coupled to a single-mode fiber (green line, PMJ-3S3AF-850-5/125-3-3-1, OzOptics, Canada) via a fiber coupler (FC, HPUCO-23-780-P-11AS, OzOptics). The output of the fiber was collimated by another fiber out-coupler and expanded using a 10× beam expander (GBE-10-B, Thorlabs, NJ). An optional neutral density filter (NC) can be inserted to adjust illumination power on the sample without having to adjust the laser output itself. Finally, the beam was further expanded and its profile homogenized using an engineered diffuser (ED, ED1-S20-MD, Thorlabs) creating a quasi-uniform square illumination pattern at the sample plane. Detection was performed by SS2 after imaging with a NIR-optimized fixed focal length (f = 50 mm, F1.8) objective lens (M5018-SW, Computar, TX) mounted on a C-mount adapter attached to the detector. The distance to the sample was set such as to obtain a magnification M = 0.097. An emission filter (FF01-893/209-25, Semrock, NY) mounted in front of the objective lens rejects scattered laser light.

Fig. 1:

Experimental setup. A: Schematic (not to scale) showing the main components. B: Photograph of the light-tight box enclosing the imaging part of the setup. Scale bar represents 10 cm. See text for a detailed description.

SS2 is mounted on a PCB (bottom green rectangle), itself connected to an intermediate PCB (middle green rectangle) hosting power and SYNC connections. A FPGA PCB (XEM7360-K160T, Opal Kelly) is connected to this board and interfaces with the host computer via a USB3 cable. Synchronization between laser and detector was obtained by feeding the SYNC output of the laser controller (NIM signal) to a pulse shaper (PS, NIM to TTL Converter, MicroPhoton Device, Italy) converting it to a TTL signal, and then to a pulse delay generator used as a frequency divider (FD, Tombak, Laser Lab Source Corporation, MT).

2.2. Samples

The sample studied in this work consisted in serial dilutions of two dyes, ICG (IR-125, Exciton, OH) and IRDye 800CW (929-70020, Li-Cor) in DMSO. 200 μL of each solution was deposited in two rows of adjacent wells in a 96 well-plate, 7 × 4 wells fitting within SS2’s field of view. Final concentrations for ICG were estimated as 1.1, 0.34, 0.11 mM, 34, 11 and 3.4 μM, and for IRDye 800CW as 70, 21, 7, 2.1, 0.7 and, 0.21 μM. An additional well containing DMSO only completed each row. Each row of sample was separated from the next by empty wells in order to avoid signal crosstalk between the two samples.

Instrument response function (IRF) datasets were separately acquired under the same conditions as the samples, using a piece of white filter paper illuminated with the laser attenuated by a neutral density filter chosen to avoid detector saturation, and the emission filter removed to allow detection of the scattered laser light. Because of the short acquisition time used for sample acquisition, the resulting IRF datasets exhibited visible shot noise and a new series of IRF acquisition was performed on a different setup at RPI (where SS2 had been moved in the meantime) using a Ti:Sa laser (Mai Tai HP, Spectra-Physics, CA) characterized by a repetition rate f = 79 MHz, used with a frequency division factor of 4 to match the synchronization capabilities of SS2.

Detector background datasets were acquired with the same settings as sample datasets by blocking the laser excitation.

2.3. Data acquisition

Data acquisition was performed using SwissSPAD Live, a custom software written in Lab VIEW (NI, Austin, TX) [4], which allows the user to control gate number G, width W, step size δt and integration time t_int, uploads the FPGA firmware and starts the acquisition process. Data was transferred asynchronously via USB3 connection, decoded by SwissSPAD Live, displayed as an intensity image and saved in an open source file format based on the HDF5 format [5].

Three series of acquisition were performed, with each series consisting of 7 measurements characterized by a different gate width W (W = 10.7, 11.7, 13.1, 15.5, 17.9, 20.3 or 22.7 ns). The series differed from one another by the laser repetition rate f_L (f_L = 20. 40 or 80 MHz). In order to maintain the required 20 MHz SYNC signal frequency, the corresponding frequency division factors were 1, 2 or 4 respectively.

All datasets were acquired with the same acquisition time and gate step, as discussed next. SS2 is capable of gate opening/closing cycles at a 20 MHz repetition rate and each frame data can be readout every 10 μs, where a frame consists of a 1-bit counter value per SPAD. The 1-bit counter records the number of photons (up to 1) detected during one of the gates opening, over a user-specified integration time t_int = (8n_G − 1)T_Sync . T_Sync = 50 ns is the inverse of the repetition rate and n_G is a parameter specifying the number of gates opened at intervals T_Sync during integration. 1-bit frames are accumulated in the memory of a FPGA controlling the camera function, to form a 8-bit gate image. In these measurements, n_G = 25 was used, and 4 gate images were summed up to form a “10-bit” gate image (max value: 4×255 = 1,020). This results in an integration time per gate image of t_int 1,020 × 10 μs = 10 ms. The gate location of each gate image was shifted by 4 times the minimum step supported by SS2 (1/56 ns), δt = 1/14 ns = 71.43 ps, 700 such steps covering the SYNC period T_Sync. The typical size of a file comprising 700 such 10-bit gate images was 40-80 MB, the range being due to the variable level of data compression.

2.4. Data Analysis

Data analysis was performed using AlliGator, a free software developed in LabVIEW, as described in previous publications [3, 6, 7]. Briefly, each file was first corrected for pile-up effect and the corresponding pile-up corrected background subtracted. In this specific study, datasets were additionally binned 2×2 to increase SNR and decays were folded onto themselves n times in order to obtain a single laser period worth of data (n = 4 for f = 80 MHz, n = 2 for f = 40 MHz, no folding for f = 20 MHz), further increasing the SNR.

Regions of interest (ROI) corresponding to the illuminated parts of each well were defined interactively and stored for reuse with all datasets. Data for each pixel was processed individually (phasor and NLSF, see below), and analyzed per ROI to obtain statistics for each dye concentration as a function of experimental conditions (laser repetition rate f_L, gate width W).

Phasor analysis was performed as described previously using AlliGator. Briefly, the IRF dataset obtained with the same parameter as the sample dataset was loaded, and the phasor of each pixel in the selected ROIs computed. These phasors values were then used as calibration phasors associated with lifetime τ_IRF = 0 ns. After loading the sample dataset, each pixel phasor was calibrated using the previously stored calibration phasor. Phase lifetime τ_φ was computed for each calibrated phasor using the standard formula:

$${\tau }_{\phi }=\frac{1}{2\pi f}\frac{s}{g},$$ (1)

where f = 80 MHz is the phasor frequency and (s, g) are the phasor components. The standard deviation of τ_φ, σ_τφ, was then obtained for each ROI.

Single-exponential decay NLSF analysis was performed as described previously using AlliGator [3]. Briefly, the IRF dataset obtained with the same parameter as the sample dataset was loaded, and the decay at each pixel in the selected ROIs was normalized and stored for further use as periodic convolution kernel. After loading the sample dataset, each pixel’s decay was fitted to a single-exponential decay convolved with the local IRF, plus a baseline. The resulting fit parameters (baseline B, amplitude A, lifetime τ), their uncertainties as well as coefficient of determination R² and reduced χ² were saved as maps for further processing. The mean lifetime τ_NLSF and standard deviation σ_τNLSF, was then obtained for each ROI.

Analysis logs (notebooks) and intermediate results were saved and further analyzed using OriginPro 9.1.

3. RESULTS AND DISCUSSION

This work was designed to answer three questions: (i) can laser repetition rates f_L larger than the supported SYNC rate of SS2 be used? (ii) Can short lifetimes be measured with SS2? and (iii) What is the influence of acquisition settings on these measurements? We will examine each question successively in the following subsections, limiting ourselves to the effect of gate width and laser repetition rate as far as the last one is concerned.

3.1. Using SS2 with laser repetition rates of 40 and 80 MHz

SS2 was designed to work in combination with a 20 MHz repetition rate laser. In order to use higher repetition rates, while still providing a 20 MHz SYNC signal requires decimating the laser signal. This is the role of the pulse delay generator (Tombak) used in our setup, which works as a frequency divider. Therefore, using a 40 MHz laser, 2 laser pulses are generated for each SYNC pulse, and similarly, using an 80 MHz laser, 4 laser pulses are generated for each SYNC pulse. In other words, when recording data for a full “SS2 period” (T_SYNC = 50 ns), we expect to observe a single decay/IRF per record when using f_L = 20 MHz, two when using f_L = 40 MHz and four when using f_L = 80 MHz. As shown in Fig. 2, which represents the normalized full field of view signal recorded with the laser scattered off a piece of white paper (IRF datasets), this is indeed the case, with the caveat that peculiar effects are created due to the large duration of SS2’s gate.

Fig. 2:

Effect of laser frequency f_L and gate width W on SS2’s IRF. Laser light scattered of a piece of white paper was used to record the setup (laser+electronics+SS2) IRF for different laser repetition rates (f_L = 20, 40 and 80 MHz), using each of the seven supported gate widths (W = 10.7, 11.7, 13.1, 15.5, 17.9, 20.3 or 22.7 ns). It is clear that for the longest widths, the “space” between gates is markedly reduced at 40 MHz, and disappears at 80 MHz, resulting in an offset corresponding to a full period integration.

For instance, Fig. 2B shows that for W = 22.7 ns, the time during which features of a non-IRF decay would be observable (which in other IRFs corresponds to times during which the signal is minimal) is limited to a very short duration of 25 - 22.7 ns = 2.3 ns per 25 ns laser period. More strikingly, Fig. 2C shows that for f_L = 80 MHz, apparently only W = 10.7 ns and W = 11.7 ns result in a window of time during which a decay would be observable, although for W = 11.7 ns, the IRF signal never reaches the baseline level, indicating that such a decay would be most likely severely deformed. For any other larger gate width, because W > T_L, the recorded signal would be integrated over a full laser period T_L, and only show non-trivial features for a duration W - T_L, as illustrated by the IRF decays in Fig. 2C, whose “baseline” is equal to 0.5 instead of ~0.

The question therefore arises whether it is even possible to measure useful decays in those conditions, and if so, what lifetime information can be extracted. A first qualitative response is provided by Fig. 3, which shows decays recorded from one of the sample’s wells in these different situations.

Fig. 3:

Effect of laser frequency f_L and gate width W on fluorescence decays recorded with SS2. Compare those curves with those of Fig. 2. A: a standard fluorescence decay curve is recognizable where the right edge of the IRF decay is located in Fig. 2A. A “mirrored” decay is visible where the left edge of the IRF decay is located in Fig. 2A. B, C: As noted in Fig. 2B and 2C, additional effects are visible at 40 MHz and 80 MHz, as discussed in the text.

Fig. 3A, corresponding to f_L = 20 MHz, looks remarkably similar to “standard” decays obtained with narrow IRFs, when focusing on the right edge of the IRF decays (compare with Fig. 2A). The left edge meanwhile exhibits a “mirror image” of the decay. These features are easily understandable when referring to the theoretical expression for a single-exponential decay convolved with a square gate (see ref. [8]for details) and comparable to other longer lifetime decays acquired with SS2 [1, 2].

Fig. 3B (f_L = 40 MHz) on the other hand confirms that the decay appears truncated for the longest gate widths (the decay doesn’t reach zero, as is clearly visible for W = 22.7 ns, but is in fact the case for all W > 15 ns). As anticipated, the situation is even more complex for f_L = 80 MHz (Fig. 3C), where not only are the decays not reaching the baseline, but in some cases, the mirror image itself does not reach the maximum value either. This is particularly striking for W = 13.1 ns, where a full decay (with reduced contrast due to the full period integration) appears visible, but the mirror image is reduced to a sharp rising part (in fact, the initial T_L − W = 0.6 ns of the mirrored decay).

Based on this qualitative assessment, it is clear that parts of all these curves contain recognizable information about the underlying single-exponential decay. The next section will address the question of whether phasor analysis and NLSF can recover it quantitatively in all cases.

3.2. Measuring short lifetimes with SS2

To get a sense of the influence of the different acquisition parameters on lifetime extraction, we focus on a single region of interest corresponding to [ICG] = 1.1 mM (top right well in the inset of Fig. 4A). Despite the very different aspects of IRFs and decays (see Fig. 2 & 3), calibration with the IRF used as 0-lifetime reference results in phasors (calculated with a common frequency f = 80 MHz) which are all very close to the 1 ns point on the universal circle (Fig. 4A, τ_φ = 0.99 ± 0.04 ns). Likewise, single-exponential fits to the observed decay result in lifetimes clustered around 1 ns (τ_φ = 1.03 ± 0.05 ns), although in general slightly larger than the corresponding phase lifetime (Fig. 4B).

Fig. 4:

Single region of interest FLI analysis. A. Phasor analysis of a single ICG well ROI (top right in the inset) as a function of laser frequency f_L and gate width W. The phasors, computed for frequency f = 80 MHz, are all clustered around 1 ns. Inset: Intensity image showing the sample studied here, and the selected ROI on the top right. Bar = 10 nun). B: Comparison of phase lifetimes (τ_φ) and fitted lifetimes (τ_NLSF) obtained for different acquisition parameters (laser frequency is identified by color). C: 3 NLSF analysis examples out of the 21 studied here. Blue curve: IRF, black: fluorescence decay, red: fit, green: residuals.

Interestingly, decays obtained with a higher laser frequency (f_L = 80 MHz) result in a smaller difference between the two methods, with the best results obtained for the shortest gate widths (τ_W=10.7,f=80 = 0.98 ns, τ_W=11.7,f=80 = 0.99 ns). This could potentially be due to the combination of 2 effects: higher laser frequency results in a larger fraction of the signal corresponding to the actual decay part (there are 4 copies of the decay, plus their mirror image, when f_L = 80 MHz), and there is no “wasted” photons used up in a full period integration for the two shortest gates (see Figs. 2C & 3C).

Nevertheless, fits are excellent no matter what condition is used, even in the most “exotic” ones, as shown in Fig. 4C (the fitted curves are generally indistinguishable from the actual decays, as can also be seen from the flat residuals).

These results therefore show that short lifetimes can be measured with SS2, no matter what settings are used. However, the above analysis being done at the single well level, does not qualify per se as fluorescence lifetime “imaging”. The next section will therefore look at this aspect specifically.

3.3. Effect of gate width and laser repetition rate on lifetime measurements

Analysis of fluorescence lifetime data at the single-pixel level can be challenging in general, but is made more so in the case of SS2, as each SPAD (pixel) is a distinct detector and is characterized by its own dark count rate and temporal response characteristic. This means that pixel-specific corrections need to be implemented to obtain reliable results. In practice, this requires the acquisition of a background dataset and an IRF dataset for each configuration to be used when recording actual samples. In a macroscopic imaging geometry, a mere sheet of white paper can be used to record the combined temporal response of the laser and detector, which can then be stored as IRF dataset. Once these detector-specific subtleties are taken care of, fluorescence lifetime data analysis is relatively simple when the decays are known to be single-exponential. In that case, NLSF analysis does not require very high SNR to be successful, but can be time consuming if the number of data points per decay is large (which is the case in the data discussed here, which are comprised of 700 points per decay). For this reason, we limited our analysis to f_L = 80 MHz and folded the 4 laser periods acquired during T_SYNC = 50 ns into a single one, in order to reduce the number of fitted data points to 175. Phasor analysis is not dependent on the complexity of the underlying decay (single-exponential, multi-exponential or otherwise), and is for all practical purpose, instantaneous, no matter how many data points the decays are comprised of. Phasor interpretation, however, is only trivial when dealing with single-exponential decays.

Fig. 5 shows an example of dual phasor and NLSF analysis of one of the datasets obtained during this work (f_L = 80 MHz and W = 10.7 ns). Fig. 5A represents the phasors of individual pixels, color-coded as a function of the well they are associated with. The color code is not particularly useful in this representation, as most phasors overlap with one another, or at least appear grouped in two main clusters, each corresponding to one of the dyes: a cluster around 1 ns groups phasors from wells 1-5 (ICG solutions), and another around 1.5 ns for pixels from wells 6-10 (IRDye800CW solutions). It is noteworthy that these phasors are centered around the universal circle, showing that they correspond indeed to single-exponential decays, as suggested by the analysis of the previous section.

Fig. 5:

Pixel-wise phasor & NLSF MFLI for f_L = 80 MHz and W = 10.7 ns. A: Phasor plot of the 10 wells labeled in B. Each cross in the plot represents the phasor of one pixel. Phasors are colored with a different color for each ROI. ICG data (ROIs 1-5) cluster around 1 ns, while IRDye800CW data cluster around 1.5 ns. B: Intensity (top) and fitted lifetime (bottom) maps. C & D: Box plots summarizing the data shown in A & B respectively. The ICG lifetimes are independent from concentration (increasing from left to right), while IRDye800CW lifetimes increase with concentration.

The NLSF results cannot be represented in the convenient graphical way offered by the phasor plot, but can conventionally be color-coded and represented as a map, as shown in Fig. 5B (bottom). Note that only a fraction of the ROIs’ pixels was used, as it turns out that each well is divided in two regions (visible in the intensity image shown in the top of Fig. 5B as a light and a bright zone), each characterized by a different lifetime. This difference is due to the angle of incidence of the laser illumination, which creates a shadow region in each well, within which autofluorescence of the well dominates the signal and skews the measured lifetime. Analysis was therefore limited to the brighter regions, corresponding to signal dominated by the dye fluorescence.

As observed in the phasor analysis of Fig. 5A, fitted lifetimes are grouped into two clusters (color-coded as indicated in Fig. 5B), corresponding to the ICG row (τ_NLSF ~ 1 ns) and to the IRDye800CW row (τ_NLSF ~ 1.5 ns). A quantitative analysis of the phase lifetimes obtained from Fig. 5A and of the distribution of fitted lifetimes of Fig. 5B (box plots of Fig. 5C & D) shows that ICG’s lifetime in DMSO (wells 1-5) is independent from concentration in the range ~10 μM – 1 mM studied here, while IRDye800CW’s lifetime in DMSO (wells 6-10) increases from 1.3 ns to 1.4 ns when concentration increases from 1 μM to 100 μM. Interestingly, the phase lifetime dispersion appears slightly larger than that of the fitted lifetime, and a slight positive bias is observed for the fitted lifetime, as was noted in the coarser analysis of Fig. 4.

Part of the observed dispersion is due to shot noise in the sample datasets, but some of it comes from shot noise in the IRF datasets. This is demonstrated in Fig. 6, which shows the result of the analysis of the same dataset discussed in Fig. 5, first with a “low” SNR IRF dataset used in Fig. 5 (black data points) and next with a “high” SNR IRF dataset (red data points). While the actual lifetimes (whether fitted or obtained by phasor analysis) are barely distinguishable in both cases, their dispersion is increased when the IRF’s SNR is lower (Fig. 6B), as expected from shot noise effects. Obviously, a similar effect would be observed when using samples with different SNR, but such an option is not always available. By contrast, obtaining a high SNR IRF is not particularly challenging and such a dataset can be reused for multiple sample analyses.

Fig. 6:

Effect of IRF SNR on lifetime dispersion. A: Distribution of fitted lifetime (τ_NLSF) versus phase lifetime (τ_φ) for the 10 different ROIs, using either a “low” SNR IRF dataset (black points) or a “high” SNR IRF dataset (red points). Error bars represent the standard deviation of each distribution. The lifetimes to not depend significantly on the IRF SNR, however their dispersion does. This is clearly demonstrated in B, which shows the standard deviation of each distribution (στ_NLSF versus στ_φ): the dispersion doubles from high to low SNR.

4. CONCLUSION

We have studied the ability of SwissSPAD2, a time-gated SPAD camera characterized by very large gates (10.7 ns to 22.7 ns), to acquire data from short lifetime NIR dyes with laser repetition rates equal to, or twice and four times larger than its nominal synchronization rate of 20 MHz Despite some deformations of the resulting decays in some combinations of laser frequency and gate width, all datasets were perfectly analyzable with the phasor approach or nonlinear least square fit, providing similar results in all cases. Data analyzed at the single-pixel level showed good uniformity within samples, and minute changes were clearly measured when sample conditions varied, despite the significant non-uniformity of the detector’s response. This study therefore establishes the value of SwissSPAD2 as a detector for NIR macroscopic fluorescence lifetime imaging.