Ultra-low photodamage three-photon microscopy assisted by neural network for monitoring regenerative myogenesis
Three-photon microscopy (3PM) enables high-resolution three-dimensional (3D) imaging in deeply situated and highly scattering biological specimens, facilitating precise characterization of biological morphology and cellular-level physiology in vivo. However, the use of fluorescent probes with relatively low three-photon absorption cross-sections necessitates high-peak-power lasers for excitation, which poses inherent risks of light-induced damage. Additionally, the low repetition frequency of these lasers prolongs scanning time per pixel, hampering imaging speed and exacerbating the potential for photodamage. Such limitations hinder the application of 3PM in studying vulnerable tissues, including muscle regeneration. To address this critical issue, we developed the Multi-Scale Attention Denoising Network (MSAD-Net), a precise and versatile denoising network suitable for diverse structures and varying noise levels. Our network enables the use of lower excitation power (1/4-1/2 of the common power) and shorter scanning time (1/6-1/4 of the common time) in 3PM while preserving image quality and tissue integrity. It achieves an impressive structural similarity index (SSIM) of up to 0.9932 and an incredibly fast inference time of just 80 milliseconds per frame which ensured both high fidelity and practicality for downstream applications. By utilizing MSAD-Net-assisted imaging, we comprehensively characterize the biological morphology and functionality of muscle regeneration processes through deep in vivo five-channel imaging under extremely low excitation power and short scanning time, while maintaining a high signal-to-background ratio (SBR) and excellent axial spatial resolution. Furthermore, we conducted high axial-resolution dynamic imaging of vascular microcirculation, macrophages, and ghost fibers. Our findings provide a deeper understanding of the mechanisms underlying muscle regeneration at the cellular and tissue levels.