Deep structural brain imaging via computational three-photon microscopy.

Abstract

Significance: High-resolution optical imaging at significant depths is challenging due to scattering, which impairs image quality in living matter with complex structures. We address the need for improved imaging techniques in deep tissues.

Aim: We aim to develop a computational deep three-photon microscopy (3PM) method that enhances image quality without compromising acquisition speed, increasing excitation power, or adding extra optical components.

Approach: We introduce a method called low-rank diffusion model (LRDM)-3PM, which utilizes customized aggregation-induced emission nanoprobes and self-supervised deep learning. This approach leverages superficial information from three-dimensional (3D) images to compensate for scattering and structured noise from the imaging system.

Results: LRDM-3PM achieves a remarkable signal-to-background ratio above 100 even at depths of 1.5 mm, enabling the imaging of the hippocampus in live mouse brains. It integrates with a multiparametric analysis platform for resolving morpho-structural features of brain vasculature in a completely 3D manner, accurately recognizing distinct brain regions.

Conclusions: LRDM-3PM demonstrates the potential for minimally invasive in vivo imaging and analysis, offering a significant advancement in the field of deep tissue imaging by maintaining high-resolution quality at unprecedented depths.