Upconversion Nanoparticle-Guided Virtual Deformable Mirror for Computational Adaptive Optics in Scattering Fluorescence Microscopy

Fluorescence microscopy in deep tissue is strongly degraded by optical aberrations, leading to reduced signal-to-noise ratio and spatial resolution. Conventional adaptive optics (AO) relies on physical wavefront-modulation hardware and iterative correction procedures, which increase system complexity and, under photon-limited deep-tissue conditions, constrain temporal resolution. We present a deep-learning-based computational adaptive optics framework, termed virtual deformable mirror AO (VDM-AO), that enables fully digital aberration correction without physical wavefront-modulation hardware. Central to this approach are dual-near-infrared lanthanide-doped upconversion nanoparticles that function as embedded guide stars for aberration sensing in scattering tissue. By integrating Zernike-based aberration modeling with a residual channel attention network, VDM-AO predicts 28 Zernike modes and digitally reconstructs aberration-corrected images from heavily distorted inputs. This strategy accurately recovers severely distorted images and achieves high-resolution imaging at depths up to 360 μm. By introducing UCNP guide stars into computational AO, this approach provides a low-cost, high-throughput solution for reliable deep-tissue aberration correction.