A. Instrumentation

Figure 1 shows the schematic of the FD-FLIM system. The sample was excited by a digitally modulated 375 nm CW diode laser (Toptica iBeam Smart 375). The laser was modulated by an auxiliary digital output port on FPGA1 (National Instruments NI 5772 digitizer and PXIe 7966 FPGA combination). The 375 nm excitation light was collimated with L1 (Thorlabs A240TM-A), reflected off a dichroic mirror (DM1, Semrock, FF376-Di01), and scanned across the sample using a pair of galvanometer mirrors (Cambridge Tech. 6220H) and an objective lens L2 (Edmund Optics 64-837, 45 mm EFL, MgF₂ coated, UV–NIR corrected triplet). The 1/e beam width at the objective focal plane was measured to be 18.5 µm with a Thorlabs beam profiler (BP104-VIS). The excitation power at the sample position was 20 mW when the 375 nm laser was operating in CW mode. During imaging, an average duty cycle between 60% and 73% and a pixel rate of 12.5 kHz were used, resulting in 0.96 µJ–1.17 µJ of energy per pixel.

The maximum permissible exposure (MPE) for skin was calculated according to the guidelines for the American National Standards Institute (ANSI) for Safe Use of Lasers.¹⁵ Both thermal and photochemical MPEs were considered. Using the 12.5 kHz pixel rate, the thermal MPE was calculated as 0.56 × 0.272^0.25 = 405 mJ/cm², where 0.272 was the imaging time in seconds, assuming no flyback dead time. Any flyback dead time would increase the total imaging time without increasing the exposure, making the assumption of no dead time the most conservative case. The photochemical MPE under the system exposure conditions was 1.0 J/cm². The exposure over a 3.5 mm limiting aperture was calculated by multiplying the energy per pixel by the number of pixels in the 3.5 mm aperture assuming 60 µm pixel spacing. The number of pixels in the 3.5 mm aperture was π3.5e−3/22/[π((60e−6)/2)2]=3403. The exposures over the limiting 3.5 mm aperture for the duty cycle ranges described earlier were 33.3 mJ/cm²–41.6 mJ/cm², approximately an order of magnitude below both the thermal and photochemical MPEs.

Images were acquired by voltage stepping the galvanometer mirrors for each pixel in a raster scan pattern. The fluorescence emission was collected by the objective lens L2, transmitted through DM1, coupled into a 200 µm core multi-mode fiber (Thorlabs—M25L02-200) using L3 (1 inch diameter lens with a focal length of 50 mm), collimated by L4 (Thorlabs F220SMA-A), and separated into three emission bands by DM2-4 and BPF1-3 (DM2—Chroma ZT405rdc, DM3—Chroma ZT488rdc, DM4—Chroma T560LPXR, BPF1—Chroma ZET405/20x, BPF2—Semrock FF01-440/40, and BPF3—Semrock FF03-525/50). The resulting emission bands were 405/20 nm, 440/40 nm, and 525/50 nm (center/FWHM). Each emission band was coupled using L4 into a 200 µm multimode fiber and directed onto a fixed gain APD (Hamamatsu C12702-11). The output of each multi-mode fiber was placed in close proximity (<1 mm) to the active area of each APD using a custom adapter plate, and thus, no coupling lens was needed between each multi-mode fiber and APD. The gain of each APD was set by turning an onboard potentiometer so that the peak-to-peak output fluorescence emission from a human coronary artery segment occupied between 1/3 and 1/2 of the dynamic range of the digitizer. This allowed for the acquisition of multiple images with the same APD gains and system calibration factors (calibration discussed in Sec. III B) while avoiding digitizer saturation at the expense of sub-optimal dynamic range utilization. The output of each APD was connected in series to two fixed gain amplifiers (Minicircuits ZFL-500LN+) and one low pass filter (Minicircuits BLP-90+). The output of each low pass filter was connected to an analog input channel on FPGA2 (National Instruments NI 5761 digitizer and PXIe 7962 FPGA combination). All APDs and amplifiers were powered with one Agilent E3630A power supply.

Figure 2 describes the FPGA synchronization scheme. FPGA1 outputted a digital pulse train (denoted as the digital modulation signal) to the digital input of the laser. Modulation frequencies were generated on FPGA1 using counters implemented in general purpose FPGA fabric. The frequency, duty cycle, and delay for each frequency were set by resetting the counter at a specific count, comparing the current count to a fixed value, and starting the counter at a fixed value, respectively. FPGA1 also outputted a trigger and a 10 MHz reference signal to FPGA2. All outputs of FPGA1 (10 MHz reference, trigger, and digital modulation signal) were generated in the same 250 MHz clock domain. All outputs were routed out of from FPGA1 using the auxiliary input/output connector of the NI 5772 adapter module through a screw terminal block (National Instruments SCB-19). Each ADC on FPGA2 was configured to lock its internal sample clock (operating at 250 MHz) to the external 10 MHz reference from FPGA1. A feedback signal was sent through the PXIe chassis from FPGA2 to FPGA1, as shown in Fig. 2. This allowed FPGA2 to disable the digital modulation signal outputted from FPGA1 when data were not being acquired to limit exposure of the sample by the 375 nm diode laser.