Deep learning enhanced dual-mode fluorescence cooperative imaging using upconversion nanoparticles.

Abstract

Multiphoton microscopy (MPM) has profoundly advanced deep-tissue imaging with superior optical sectioning capabilities. However, achieving high-resolution imaging at significant depths remains a challenge due to light scattering and resolution-penetration trade-offs. Here, we present a deep learning enhanced dual-modal fluorescence cooperative imaging (DL-DMFC) approach to achieve deep-penetration high-resolution imaging. By utilizing the multiple long-lived intermediate states, lanthanide upconversion nanoparticles (UCNPs) simultaneously induce two-photon (λemission1 = 808 nm) with higher penetration and four-photon (λemission2 = 455 nm) fluorescence with higher resolution under a single 980 nm pump source excitation. To synergistically leverage the advantages of imaging at two fluorescence, we trained artificial neural networks incorporating a dual mechanism based on adversarial training with cyclic consistency constraints is employed to establish a cross-domain mapping between the dual-modal signals. We demonstrate that this synergistic excitation and computational framework enable high-resolution (51% transverse resolution enhancement), anti-scattered 3D imaging beyond 500 μm. This approach solves the problem of penetration-resolution trade-off in MPM and provides a new strategy for deep tissue thick scattering imaging.