Interleaving SLIDE microscopy

Two-Photon microscopy (TPM) is an intravital imaging technique with deep tissue penetration, high three-dimensional resolution and low photobleaching. However, imaging speeds have traditionally been limited because images are generated by point scanning. Recently, Spectro-Temporal Laser Imaging by Diffractive Excitation (SLIDE) introduced a high-speed laser-scanning approach for two-photon microscopy, harnessing a swept-source laser in combination with a diffraction grating for inertia-free point scanning with line-rates at Megahertz (MHz) frequencies and frame-rates in the kilohertz range. Notably, the pixel-rate reached up to Gigapixels/s, which scales with ultra-short second harmonic generation signals. However, this nanosecond pixel dwell time is too short for fluorescence decays. Here, we present an adaptive approach to high-speed multiphoton imaging where the pixel-rate of the SLIDE microscope can be adapted to the fluorescence lifetime of the sample. This is achieved by interleaved sampling, allowing for larger temporal intervals in the time domain while preserving spatial sampling density. We demonstrate adaptive frame-rate SLIDE microscopy by flexible, active pulse modulation on an electro-optic modulator (EOM) and discuss how this can be extended to future smart SLIDE systems where the pixel dwell time can be optimally set for each sample.

Two-Photon Microscopy (TPM) (Denk et al., 1990) has become an indispensable tool for deep tissue optical imaging due to its intrinsic optical sectioning, reduced photobleaching, and enhanced imaging depth in scattering tissue. However, the achievable frame-rates in conventional TPM are limited by the mechanical inertia of the beam scanning systems. Thus, while TPM is routinely used in neuroscience for studying calcium-based neuronal activity imaging (Chen et al., 2011; Lin and Schnitzer, 2016), the rapid dynamics of voltage-based neuronal activity require higher frame-rates in the kilohertz range (Katona et al., 2012; Kazemipour et al., 2019; Wu et al., 2020).

Spectro-temporal laser imaging by diffractive excitation (SLIDE) (Karpf et al., 2020) enables ultra-fast TPM by introducing a non-mechanical and ultra-fast scanning mechanism along the fast axis. SLIDE leverages a Fourier domain mode-locked (FDML) laser (Huber et al., 2006), which provides a rapidly tuneable, narrow-linewidth swept source operating in the Megahertz regime. The output of the FDML laser is modulated into picosecond pulses using an electro-optic modulator (EOM), and the resulting pulse train is amplified in a Master Oscillator Power Amplifier (MOPA) (Karpf and Jalali, 2019), achieving peak powers up to kilowatt range (Kutscher et al., 2023). By directing this laser source onto a diffraction grating, the rapid sweep rate of the FDML-MOPA produces an inertia-free x-axis scan with distinct pulse-to-pixel mapping at up to GPixel/s (Karpf et al., 2024). Each sweep generates a line scan, reaching MHz line scan rates (Karpf et al., 2024). In combination with a galvo scanner for the slow axis, this enables SLIDE to reach point-scanning imaging at kHz frame-rates (Karpf et al., 2020).

However, the high scanning rate introduces a limitation: the pixel dwell time (∼ 1 ns) is shorter than the fluorescence lifetime of typical dyes (Lakowicz et al., 1992; Suhling et al., 2015). Due to the pulse to pixel mapping, this mismatch leads to bleed-through of photons to adjacent pixels and limits the ability to obtain fluorescence lifetime imaging microscopy (FLIM) data, inherent to SLIDE (Karpf et al., 2020). This effect is visualized in Figure 1C. Here, the emission response to each excitation pulse is illustrated for different pulse repetition rates with increasing time between pulses. For high pulse repetition rates, the inter-pulse time is shorter than the fluorescence decay time, which would lead to pixel cross-talk and thus to a loss in spatial resolution for the reconstructed image. If pulse rates are lowered, comparable to conventional TPM microscopy (≤ 80 MHz) (Suhling et al., 2015; Giacomelli et al., 2015), bleed-through can be prevented or reduced.

Previous studies reported on methods to adapt the spatial sampling pattern to the sample (Katona et al., 2012; Chen et al., 2011; Lechleiter et al., 2002; Otsu et al., 2008), and even adjust temporal sampling to accommodate longer fluorescence lifetime decays (Iyer et al., 2024) for laser scanning microscopes (LSM). SLIDE has been shown to support flexible sampling patterns for adaptive acquisition (Jiang et al., 2020). Here, we employ adaptive sampling in the temporal domain to allow for longer decay times arising from fluorescence emission or to increase acquisition speed by harnessing ultra-fast second harmonic generation (SHG) signals (Karpf et al., 2024). To address this, the fully programmable spectral scanning of SLIDE enables an interleaving scheme, where subsequent sweeps are utilized to sample a whole line (cf. Figure 1B), yet leaving enough time to accommodate fluorescence decays. For this, the electro-optic pulse generator is driven by an analog input signal (see Figure 2). This enables a pulse-on-demand operation, where the pulse pattern is changed on the fly. By programming an interleaving pattern on an arbitrary waveform generator (AWG) (Karpf et al., 2015; Karpf et al., 2017), the effective pixel dwell time can be programmed. In the implementation presented here, the line scan rate is kept constant (1.6 MHz) while an image line is constructed of multiple line scans using interleaved sampling. This adaptive pixel-rate SLIDE microscope allows for high-speed fluorescence lifetime imaging microscopy (FLIM) (Karpf et al., 2020), where the acquisition speed is matched to the sample.

Laser source

While previous SLIDE systems employed home-built FDML lasers (Karpf et al., 2020; Kutscher et al., 2023; Karpf et al., 2024), the SLIDE system presented here uses a commercial FDML laser (Optores NG-FDML-1060 with custom 10–30 nm sweep range) as the seed in the FDML-MOPA configuration (Karpf and Jalali, 2019). The integrated function generator serves as synchronization signal source for pulse triggering, signal detection, and control of the newly introduced AWG card, as illustrated in Figure 2. The FDML laser and electro-optic modulator (EOM) are combined with a home-built, two-stage ytterbium-doped fiber amplifier (YDFA). This configuration forms a pulsed wavelength-swept laser source with 35 ps pulse width and capable of generating peak powers in the kilowatt range, depending on the operated pulse rate (Karpf and Jalali, 2019; Kutscher et al., 2023).

Pulse trigger switch

In previous SLIDE microscopes, a phase-locked loop (PLL) generated a clock signal for an electronic pulse generator (Sigtrona APG1780DA-SMA). This ensured a phase-stable pulse train with constant repetition rate. These pulses drive the EOM (Thorlabs LNX1020 A), creating a constant sequence of pulses phase-locked to the FDML sweep rate, where each pulse addresses one imaging pixel. In this work, the interleaved pulse pattern was generated on a reprogrammable arbitrary waveform generator (AWG) card (Spectrum Instrumentation M4i.6633), which triggers the pulse modulation electronics. A detailed schematic is shown in Figure 2. An additional HF-switch is inserted to toggle between regular PLL operation of the pulses or the AWG-pattern. The AWG card is operated via PCIe, which supports data transfer rates of up to 2.8 GB/s, enabling rapid, adaptive reprogramming of the waveform. Here, the AWG produces the required line-wise phase shifts and pulse rates for different interleaving settings, as visualized above in Figure 1B.

Interleaved mask generation

Under standard, non-interleaved operation, the EOM operates at a stable pulse rate of 820 MHz, which is fixed to the FDML sweep rate of 1.6 MHz. Correspondingly, 512 pulses are evenly distributed over each FDML sweep, equating to 512 pixels along the x-dimension. The y-scan is performed by a galvanometric mirror driven with a 1.56 kHz sine wave. The ratio of the FDML sweep rate to the y-scan rate is chosen to contain 1,024 x-scans per image, yielding an image resolution of 512 × 1024 pixels. A 512 × 512 image size would be preferable (i.e., two-fold faster frame-rates), but the galvo scanner exhibited severe hysteresis and phase instability when driven at higher frequencies and according amplitudes. Hence, 1,024 lines per image (1.56 kHz frame-rate) were employed in this study.

Interleaved raster scanning leverages the programmability of an AWG card to mitigate resolution loss when longer pixel dwell times are required. To achieve an interleaving factor of 2, the pulse repetition rate is effectively halved and the pulse train is shifted by a phase of π (half a Pulse Interval) after every sweep (x-scan), thus maintaining an effective average pulse rate of 410 MHz. A similar time-encoding was previously employed to stimulated Raman scattering (SRS) microscopy using FDML lasers (TICO-Raman) (Karpf et al., 2015). The intermediate pixel pattern (see Figure 1B) allows to bin two successive line scans and reconstruct them as a single line. Due to the line-wise phase shift, the merged scans effectively replicate a single high-density scan acquired at the original pulse interval (512 pulses per image line). Thus, a simple binning algorithm can be used, where subsequent lines are averaged. This binning operation reduces the initial 512 × 1024 image, obtained at the base pulse rate f 0 = 820 MHz, to a final resolution of 512 × 512 pixels.

This process can be generalized for any interleaving factor F = 2 n, where the corresponding pulse rate is given by f rep = f rep, 0 / F. The phase shift per line scan is ϕ = 2 π / F, which requires an equivalent increase in the number of line scans to reconstruct and bin a single output line. To maintain the target 512 × 512 image dimensions, both the y-scan rate of the galvanometric mirror and the overall imaging frame-rate are reduced proportionally to the interleaving factor.

In Table 1 we summarize the interleaved patterns employed with the corresponding imaging rates, output dimensions, and applied laser powers.

Volumetric imaging

Axial sectioning for volumetric imaging is achieved with a piezo-driven objective scanner (Thorlabs PFM450E), which translates the microscope objective along the optical axis, shifting the focal plane through the sample. Thereby, volumes can be acquired by moving the focal plane over an axial range. Like the galvanometric mirror, the objective scanner is scanned with a sine frequency locked to the sweep rate of the FDML laser. Following this, the volume-rate (3D) is defined by the quotient of frame-rate (2D) and the amount of acquired layers per volume. As interleaving reduces the two dimensional frame-rate, the volume rate is reduced accordingly.

Detection

To analyze fluorescence lifetimes or SHG signals, the SLIDE system is operated with a high-speed hybrid photo detector (HPD, Hamamatsu R11322U-40). The rise time of 400 ps is significantly shorter than the fluorescence lifetimes of conventional dyes (typically nanoseconds (Lakowicz, 2006; Giacomelli et al., 2015)), which is a prerequisite for temporally resolving the fluorescence decay. Due to its relatively low gain, the HPD is connected to a transimpedance amplifier.

To resolve the fluorescence contrast, we use a 400 MHz transimpedance amplifier (TIA, FEMTO HCA-400M-5K-C). This TIA is able to resolve the expected fluorescence lifetimes, acquired with single pulse FLIM (Eibl et al., 2017), while operating with a high gain of 5 × 1 0 3 V/A. Due to the low photon counts, the high amplification ensures the use of the full dynamic range of the operated digitizer (± 400 mV).

Due to the fast response of SHG, the operated amplifier needs to match the 400 ps rise-time of the HPD to reduce temporal broadening of the SHG pulses. For this a 2.5 GHz transimpedance amplifier (TIA, Sigtrona DCWBA3500M) was employed. For SHG recordings the TIA is operated with a lower gain of 1.5 × 1 0 3 V/A, allowing for higher dynamic range of the signal output. Thereby, a saturation of the ADC can be prevented while increasing the signal through higher laser peak powers, an opposite requirement to the fluorescence decay based application.

In both cases, the amplified analog signal is then fed into a 4 GS/s analog-to-digital converter card (ADC, AlazarTech ATS9373), satisfying the Nyquist-Shannon sampling criterion (Shannon, 1949; Nyquist, 1928).

Second harmonic imaging

To overcome the pulse repetition rate limitations posed by fluorescence lifetimes, SLIDE can also employ harmonic imaging. SHG occurs in non-centrosymmetric materials and is commonly utilized in tissue imaging, for example, in imaging collagen fibers (Pijanka et al., 2019; Ambekar et al., 2012). Two advantages of SHG are the high, non-saturating signal intensity and the quasi-instantaneous signal response. In case of 35 ps excitation pulse length, the SHG signal duration lies in the picosecond range. The effective instrument response function (IRF) is ultimately governed by the systems’ detection bandwidth.

In case of the faster TIA with 2.5 GHz bandwidth, the maximum detection bandwidth is limited by the HPD response (600 ps FWHM), allowing for 1.6 Gigapixels/s acquisition (see Figure 5). This is an improvement over previous SLIDE systems, where the 400 MHz TIA limited the pixel dwell time to 1 ns (Karpf et al., 2024).

Fluorescence decay reconstruction

In contrast to time correlated single photon counting, where the temporal distribution of single photon events for multiple pulses is measured for a pixelwise decay fit, we are operating the detection in an analogue mode for single pulse FLIM operation (Eibl et al., 2017). By exciting the sample with a longer pulse length, multiple fluorophores are excited at once, requiring fewer pulses to generate a sufficient amount of photons for a fluorescence decay fit. In combination with a high speed ADC (4 GS/s), the temporal decay curve of an ensemble of multiple photons can be determined. The phase locking of FDML (x-scan), galvanometric mirror (y-scan) and AWG (Pulse clock) allows for a pixel wise pulse mapping, ensuring a pixel distinct remapping of the fluorescence signal.

For decay contrast image reconstruction, the open source ImageJ expansion Fiji (Schindelin et al., 2012) is used. Therefore, a time stack [t,y,x] is fed into the plugin FLIMJ (Gao et al., 2020) and filtered with a 5x5 regional binning (i.e., a 2-neighbor window, summing counts over the central pixel and its surrounding pixels) to increase the amount of available photon counts per pixel. Afterwards, a pixel wise decay fit is operated to reconstruct a tau mask including the calculated decay times for every pixel. This tau mask is then combined with the raw non binned image [y,x] into a HSV depiction. The shown decay plots were generated and fitted in Origin 2025 (Dataset OriginLab Corporation, 2025) for the specified pixel.

Microscopy setup

For TPM, the 2D laser scan is coupled into an inverted microscope (Nikon Eclipse Ti-E) and focused onto the sample by a water immersion objective with cover glass correction (Nikon ×40/1.15 WI 0.6 WD). The field-of-view (FOV) was measured to be 150 μ m x 140 μ m.

Samples

In this work, three measurements of different samples are presented to evaluate various effects of the interleaving method on signal response, spatial resolution and fluorescence decay analysis.

The fluorescent beads (Thermo Fisher F8825), which were resolved in a water solution, are used to show that image resolution is retained for interleaved sampling, as well as for analysis of pixel cross-talk and demonstration of the interleaving pulse pattern in the temporal domain.

To visualize the increase in laser pulse power through the reduction of pulse rates and the maximal resolvable pulse rate for the presented imaging system, SHG images of Urea crystals are shown. The non saturating nature of SHG allows a direct conclusion on signal increases through higher laser peak power.

To test the ability to resolve different decay characteristics, paraffin embedded histological slices were labeled with different fluorescent dyes and imaged. Histological examinations were approved by the Ethics Committee of the University of Lübeck (No. 2024589).

For volumetric imaging, living and highly motile Euglena gracilis microalgae cells are used. For this purpose, a sparse cell culture is first centrifuged and then placed between a microscope slide and a cover slip, separated by a 1 mm spacer. Thereby providing enough space for the cells, to move in all three dimensions unhindered. As the Euglena gracilis cells provide autofluorescence from their chloroplasts, they can be imaged without any staining.

Tissue staining

To remove paraffin from the slides for the staining process, the samples were immersed twice for 10 minutes in xylol and rehydrated with a decreasing ethanol concentration (e.g., 100%, 90%, 70%) and finally rinsed in 0.9% sodium chloride. To stain similar cellular structures as in Hematoxylin and Eosin (H&E) staining, the fluorescent dye SYTO82 (Thermo Electron Karlsruhe GmbH, Karlsruhe, Germany) was used to stain nuclei and Rhodamin 6G (Carl Roth GmbH + Co. KG, Karlsruhe, Germany) for non-specific staining of cellular structures. The sample was incubated for 15-min in a 10 μ M Rhodamin 6G solution, followed by a short rinse with water and another 15-min incubation in 3.14 μ M SYTO82.

Owing to the large field of view, a sub-resolution bead would be sampled by only a single pixel, which is insufficient to verify that the × 8 interleaving samples a given structure consistently across all interleaved sub-frames. We therefore used larger 2 μ m beads, which extend over several pixels allowing us to detect image artefacts arising from timing errors, jitter, or sweep inhomogeneities.

Figure 3 demonstrates the improvement in temporal resolution for fluorescence imaging enabled by interleaved imaging. To this end, fluorescent beads were imaged with two different interleaving factors. The fluorescence response signals for x2 and x8 interleaving were shown for the two upper beads (A 2, B 2). Here, the laser power was set to a level where no signal saturation was detected for the x8-interleaved modality. Thereby, for both acquisitions, the same laser pump current was applied (Table 1), leading to a comparable average power of around 130 mW on the sample, yet different pulse peak powers and hence favorable nonlinear excitation rates at lower pulse repetition rates. For comparison, the image cut-off levels of A 1 and B 1 were chosen to be the same to show the increase in brightness with interleaved imaging. Regarding temporal resolution, the time traces acquired with x2 interleaving (A 2) exhibited overlapping fluorescence signals, resulting in high pixel cross-talk. This was also visible in the plotted raw data of the interleaved acquisition (Figure 3 B 3). To achieve a better signal-to-noise ratio (SNR), the images were averaged 20-fold, leading to recording rates for the shown images of 80 Hz for x2 interleaving, 20 avg, and 20 Hz for x8 interleaving, 20avg. To analyze the mean signal over time, the digitized fluorescence signal acquired for the reconstructed image (Figure 3 A 1 B 1) was averaged and mapped to the dynamic range of the operated ADC (800 mV).

To visualize the effects of fluorescence bleed-through between adjacent pixels and the positive effect of interleaved SLIDE, fluorescence beads with a diameter of 500 nm (Thermo Fisher F8812) are imaged with non-interleaved and interleaved sampling. As the digital sampling of the laser scan (∼ 500 nm) is too low to resolve sub-resolution beads (200 nm), the lowest resolvable bead size was used. The resulting acquisition is shown in Figure 4 and large effect of the fluorescence bleed-through for non-interleaved sampling (a, e). For x8-interleaved sampling, no fluorescence bleed-through is detected.

While interleaving could accommodate long-lived fluorescence decay states, it also led to higher peak powers at the same average laser power and hence higher nonlinear excitation rates (cf. Figure 3). The pulse flexibility could also be harnessed for even faster pixel dwell times when dealing with very fast nonlinear signals. Figure 5 illustrated SHG imaging of Urea crystals using both non-interleaved and interleaved configurations, demonstrating the flexibility of SLIDE in matching repetition rates to signal dynamics. As for the fluorescence measurements in Figure 3, the SHG images (B, C, D) were acquired with a similar laser pump power, as shown in Table 1. The almost instantaneous response of SHG allowed for pixel rates of up to 1.64 GHz (A). At this pulse repetition regime, the pulse peak power is significantly reduced (comp. Table 1 only 3.2 W peak power at 183 mW average laser power on sample). To compensate for this, the pump current is increased compared to the other acquisition states, generating an average power of 717 mW on the sample. At decreased pixel rates (B-D), lower average powers (135–175 mW), compared in Table 1, were applied to the sample. The images were acquired at a frame-rate of 1,562 Hz (A-C) or 781 Hz (D) and were averaged 20 times to improve the SNR.

As a third application example, we demonstrate interleaved SLIDE with lifetime-based contrast. For this, a histological tissue section was stained with SYTO82 (nuclei) and Rhodamin 6G (unspecific) and was imaged for fluorescence lifetimes, as shown in Figure 6. The different fluorescence lifetimes were visualized using an HSV depiction. Here, the lifetime differences between nuclei and tissue were clearly distinguishable for averaged (B, 40 Hz) and unaveraged imaging (A, 400 Hz). The fluorescence decay plots were reconstructed for averaged (D) and unaveraged (C) data of the same pixel.

To visualize the transfer from 2D imaging to volumetric imaging, the piezoelectric objective scanner is operated with a sinusoidal signal with 12 Hz oscillation, resulting in a volume scan rate of 24 Hz. The 2D scan is operated under x4-interleaved SLIDE with a frame-rate of 781 Hz. This results in 32 frames per volume. This scan records the dynamic motion of Euglena gracilis cells, shown in Figure 7 and available as four dimensional tif-stack in the accompanying data (time, z, y, x). The characteristic timescales that govern these pixel cross-talk and bleed-through effects, and thus motivate the choice of interleaving factor, are summarized schematically in Figure 8.

SLIDE provides multiphoton imaging at nanosecond pixel dwell times which encounters several physical limitations, namely, a low number of photons per pixel and pixel dwell times shorter than the fluorescence decay times. This work presents an interleaving approach that allows to adjust the SLIDE imaging parameters to the temporal behavior of the sample.

By imaging micrometer-sized fluorescent beads, we showed that temporal interleaving maintains spatial resolution (cf. Figure 3). The eightfold longer inter-pulse time avoids pixel cross-talk by increasing the pixel dwell time to ∼ 10 ns, thus allowing to resolve the fluorescence decay curve (Figure 3 B 2). As a positive trade-off effect, the accompanying higher peak power makes efficient use of the quadratic excitation dependency, thus leading to much increased fluorescence signal levels (Figure 3 B 2). The mean signal level increased fivefold by using the interleaving pulse pattern. Over all interleaving factors, no jitter or inter sweep phase mismatches were observed.

The statistical analysis of the detection precision for a 500 nm bead demonstrates the primary benefit of interleaved SLIDE (Figure 4). By directly matching the pixel dwell time to the decay characteristics of the fluorescent beads, fluorescence bleed-through is prevented, increasing the precision of the two-photon image. For non-interleaved SLIDE, the fluorescence decay exceeds the pixel dwell time, so the detected bead structure is elongated in the x-direction. The increased pixel dwell time of x8-interleaving suppresses this effect and yields a circular image of the bead. The spatial phase stability of the x-scan follows directly from the phase stability of the FDML sweep (Grill et al., 2022). Also, the higher SNR of interleaved SLIDE decreases statistical noise in the fluorescence signal. This effect increases the precision of the operated gaussian fit, used to estimate the per-frame bead localization. As a result, the standard deviation of the frame wise two-dimensional Gaussian-fit is decreased from 990 nm under Non-Interleaved imaging to 186 nm under x8-Interleaved imaging (Figures 4E,J).

On the other end, the flexible pulse modulation allows to increase pixel rates when imaging quasi-instantaneous harmonic signals. In Figure 5 we show up to 1.6 GHz pixel rate imaging of SHG signals from Urea crystals, as the fast detection bandwidth allows to resolve sub-nanosecond pixel dwell times without pixel cross-talk. Here, interleaving can also be harnessed to boost signal levels, especially as harmonic imaging does not suffer from signal saturation (Figure 5 right). This demonstrates that the same microscope can be adjusted for very different signal types by adapting the pulse repetition rate to the temporal properties of the sample.

Interleaved imaging applied to biological specimen imaging was presented in Figure 6. By fitting mono-exponential decay curves to the fluorescence decay signal, a fluorescence decay based contrast was presented with a frame-rate of up to 400 Hz. Using only a single detector, the fluorescence contrast allows for clear distinction between cell nuclei stained with SYTO82 and surrounding tissue stained with Rhodamin 6G, by color-coding the fluorescence lifetimes between 0.8 and 2.5 ns. Tenfold frame averaging achieves increased SNR for high fidelity fit results (Figure 6B), while a 40 Hz recording rate still provides high temporal resolution of 25 ms per averaged image. A quantitative evaluation of single-pulse FLIM under interleaved SLIDE, including lifetime standards and a detailed fit-quality assessment, is currently the subject of ongoing work and will be reported separately.

Finally, interleaved SLIDE extends naturally to volumetric imaging (Figure 7). Here, the interleaving factor directly governs the volume-rate: since the volume-rate equals the 2D frame-rate divided by the number of axial planes (=frames), the F-fold frame-rate reduction of interleaving propagates linearly into the third dimension. Operating Euglena gracilis imaging at x4-interleaving sets the pixel dwell time (∼ 5 ns) long enough to suppress fluorescence bleed-through, while retaining a 781 Hz frame-rate that, distributed over 32 axial planes, yields a 24 Hz volume-rate across a 210 × 150 × 64 μ m 3 field of view. Interleaving thus becomes a programmable trade-off knob that balances bleed-through suppression against volumetric speed. The same temporal degree of freedom that suppressed lateral bead elongation now sets the achievable 3D imaging rate.

Over the duration of the experiments, no photo damage was detected. Especially in case of the Euglena gracilis microalgae (Figure 7), the cells did not show any effect at the applied laser powers, even at longer illumination durations. In case of x0.5-interleaving (1.64 GHz) with higher required laser powers (717 mW), the effects of this thermal load would have to be analyzed in an additional study.

The flexibility in the presented AWG-powered SLIDE system enables an adaptive approach: While we present image-wise adjustment of the interleaving pulse pattern, the technical capability of the system allows for real-time, low latency (between 5 and 10 ms) adaptation of the waveforms. As one waveform contains the control signals for a full image scan, the pulse repetition rate can be defined per pixel. Thus, in principle, individual regions of interest (ROIs) of the image can be addressed at individual, optimally chosen interleaving settings. In the future, such an approach could be combined with real-time signal analysis, allowing the microscope to automatically choose suitable pulse timing and excitation settings depending on fluorescence lifetime or signal strength. In this way, future SLIDE systems could optimize the trade-off between imaging speed, signal quality, and temporal resolution for each sample individually.

Biological specimen often show a large variety of imaging needs: Some regions of the sample have lower signal strengths, longer fluorescence lifetimes or sharper spatial features. In order to optimize imaging parameters while still providing kHz frame-rates for capturing fast dynamics, a smart microscope should be able to accommodate the best imaging parameters for each region of interest. In this manuscript, we presented a technical implementation of an adaptive SLIDE microscope that shows low latency (between 5 and 10 ms) to update the imaging parameters on-the-fly. Future developments will focus on real-time signal analysis to enable a smart microscope that adapts to the sample under investigation. Beyond imaging speed, such a smart microscope could actively balance the trade-off between signal-to-noise ratio, frame-rate, and applied laser dose, selecting for each region the operating point that provides sufficient signal at the lowest necessary excitation. Such a real-time adaptive smart microscope will help to shed light on fast biological processes with kHz multiphoton imaging.