N25 - Pan-sharpening to Achieve Computational Super-Resolution in Fluorescence Lifetime Imaging Microscopy

Introduction: : Fluorescence lifetime imaging microscopy (FLIM) is a powerful technique that maps the biochemical microenvironment of living samples by measuring the decay dynamics of fluorophores. Unfortunately, photon budget limits, detector design, and the need for rapid frame rates constrain the spatial resolution of FLIM. Hardware super-resolution techniques can address some of these limitations, but they require specialized optics [1]. On the other hand, computational approaches [2] are primarily suited for fluorescence intensity data and cannot be applied to FLIM data due to the inherent correlation between the temporal and spatial information in FLIM images. To achieve computational super-resolution in FLIM, we propose a pansharpening workflow that draws inspiration from remote sensing. In pansharpening, a high-resolution panchromatic band is fused with low-resolution multispectral bands to yield an image that is both sharp and spectrally accurate. Analogously, we treat the FLIM intensity image as a spatially rich "panchromatic" channel, and the phasor in-phase (I) and quadrature (Q) components as "spectral" channels that carry chemical information. We first apply computational super-resolution to the intensity image using the enhanced Super-Resolution Radial Fluctuations (eSRRF) algorithm [2], and then use this high-resolution intensity to sharpen the I and Q images. Crucially, spatial sharpening is applied only to the intensity layer, ensuring that the temporal information and lifetime values derived from I and Q remain numerically accurate. We validated the workflow using fixed bovine pulmonary artery endothelial cell (BPAE) FLIM images captured by our two-photon frequency-domain FLIM system [3].

Materials and

Methods: : Fixed bovine pulmonary artery endothelial (BPAE) cells (Invitrogen F36924) were labeled for: mitochondria with MitoTracker Red CMXRos, F-actin with Alexa Fluor 488 phalloidin, and nuclei with DAPI. Imaging was performed using a two-photon frequency-domain fluorescence lifetime imaging microscopy (FLIM) system, called instant-FLIM [3], which captured the in-phase (I), quadrature (Q), and intensity (DC) components for each pixel. Our FLIM pansharpening method is compatible with computational localization techniques. In this study, we employed the eSRRF [2] due to its widespread use in the literature. eSRRF leverages temporal correlation across image frames to achieve super-resolution. For each FLIM component (I, Q, DC), we collected 100 frames of 560 × 560 pixels and averaged them to improve signal-to-noise ratio, since each component image provides an unbiased estimate. To isolate spatial information in the DC channel, I and Q were transformed into g = I/DC and s = Q/DC phasor coordinates. eSRRF was then applied to the DC stack to generate an upsampled, spatially sharpened image. Pansharpening using component substitution was carried out by first upsampling g and s images using Catmull-Rom cubic spline interpolation. These were then fused with the sharpened DC image to reconstruct sharpened I and Q images. Additionally, a composite image combining the sharpened I, Q, and DC images was obtained by calculating τ = Q/Iω and mapping hue to lifetime, value to intensity, and setting saturation to 1 using the HSV colormap. This approach ensures that we do not alter the sample's lifetime information while enhancing the spatial features.

Results, Conclusions, and Discussions:: Results and

Discussion:
Fig. 1 (A-C) on the first column displays the averaged in-phase (I), quadrature (Q), and intensity (DC) from the 100-frame stacks. The second column shows the the resulting images of I, Q and DC after sharpening. We can observe that both the I and Q components exhibit sharper edges, indicating that spatial sharpening features have been integrated into the chemical information contained within I and Q. Additionally, Fig. 1 (D) demonstrates a significant difference between the original composite image and the sharpened composite. The mitochondria structure is clearly visible in the sharpened image, whereas it was not discernible in the original composite image.

Conclusion:
In conclusion, our pansharpening approach has successfully improved the spatial resolution of fluorescence lifetime imaging (FLIM) data without compromising the integrity of the lifetime information. This is the first demonstration of a technique to perform computation localization super-resolution FLIM that is capable of preserving the rich chemical information present in FLIM images. By effectively sharpening the in-phase (I) and quadrature (Q) components, we have created clearer and more distinct images, enhancing the visualization of crucial biological structures such as actin filaments. This advancement provides significant benefits for researchers and biologists who aim to enhance their analyses of FLIM images, ultimately leading to more accurate interpretations of cellular dynamics.

Acknowledgements and/or References (Optional):: [1] Oleksiievets, Nazar, et al. "Fluorescence lifetime DNA-PAINT for multiplexed super-resolution imaging of cells." Communications biology 5.1 (2022): 38.
[2] Laine, Romain F., et al. "High-fidelity 3D live-cell nanoscopy through data-driven enhanced super-resolution radial fluctuation." Nature Methods 20.12 (2023): 1949-1956.
[3] Zhang, Yide, et al. "Instant FLIM enables 4D in vivo lifetime imaging of intact and injured zebrafish and mouse brains." Optica 8.6 (2021): 885-897.