Y-I Chen, T.D. Nguyen, Y-J Chang, S-C Liao, Y-A Kuo, S. Hong, G. Rylander III, S.R. Santacruz, T. Yeh
University of Texas at Austin,
United States
Keywords: multiphoton microscope, fluorescence lifetime, retinal diagnosis
Summary:
Optical coherence tomography (OCT) and confocal scanning laser ophthalmoscope offer astounding opportunities to image the complex retinal structures. However, a noninvasive mapping of the subcellular distribution of endogenous fluorophores based on their spectra and decay profiles is lacking. Although fundus autofluorescence microscopy (FAFM) is a well-established technique that shows different patterns of FAF distribution at retina, it didn’t provide the critical information about the pathogenetic relevance to the present fluorophores. Fluorescence lifetime imaging ophthalmoscopy (FLIO) has become an important tool in retinal diagnosis as it provides additional information on the temporal characteristics of the fluorescence decay. However, the current methods to generate fluorescence lifetime images are either computationally intensive or unreliable when the number of photons acquired at each pixel is low. Here we introduce a new deep learning-based method termed flimGANE (fluorescence lifetime imaging based on Generative Adversarial Network Estimation) that can generate fast, fit-free, precise, and high-quality fluorescence lifetime images even under the photon-starved conditions. We demonstrated our model is not only up to 2,800 times faster than the gold standard time-domain maximum likelihood estimation method but also provides more accurate analysis in barcode identification, cellular structure visualization, Förster resonance energy transfer characterization, and metabolic state analysis in live cells. The further clinical application of flimGANE is in our custom-built two-photon fundus autofluorescence lifetime imaging ophthalmoscopy (2P-FAFLIO). Two-photon excitation (TPE)-based imaging overcomes the filtering of ultraviolet light by the lens of the human eye. Combining TPE with FLIO offers great potential for monitoring the metabolic transformation in the retina and detecting retinal diseases at earlier stages before irreversible structural damage has occurred. However, when applying to rabbits and non-human primate models, TPE suffers from the issue of the eyes with a low numerical aperture (0.1-0.2), which significantly reduces the autofluorescence signal and leads to the poor quality of the retina images. Integrating temporal focusing, pulse picking, and hyperspectral detector in our imaging system enables us to improve the speed and quality of FLIO imaging by generating a two-photon focal spot of 5 μm in xy and 10 μm in z. With a 5-fold bigger excitation spot while maintaining small axial confinement, we can increase the number of detected photons by 25 times. Furthermore, reducing the pulse repetition rate increases the TPE efficiency, resulting in a higher fluorescence signal for the same average excitation power. We expect a 10-fold increase in autofluorescence signal when the repetition rate of our laser is reduced from 80 MHz to 8 MHz. Additionally, with the implementation of flimGANE in our 2P-FAFLIO, we can overcome the low-photon-count issues. In sum, we anticipate the 2P-FAFLIO imaging acquisition time to be reduced from hundreds of seconds to tens of seconds, with a minimum reduction in spatial resolution. The system will be validated using in-vivo rabbit models. The acquired autofluorescence lifetime can help us understand the states of the endogenous fluorophores within the retina and their roles and changes during the development of retinal diseases.