Fluorescence-illuminated diffraction tomography using explicit neural fields

Abstract

Abstract Multimodal fluorescence and phase imaging provides complementary information about biological samples. However, existing simultaneous fluorescence-phase techniques primarily operate in transmission mode and are limited to thin specimens, restricting their use in thick tissues and in vivo applications. Although multiphoton microscopy enables deep-tissue fluorescence imaging, combining it with three-dimensional refractive-index (RI) imaging remains challenging because few methods can reconstruct the RI of bulky, label-free tissues in reflection mode at subcellular resolution. Here, we introduce fluorescence-illuminated diffraction tomography (FDT), which reconstructs the 3D RI of label-free and fluorescence-labeled objects from diffracted fluorescence images acquired in reflection mode under two-photon excitation. FDT uses the transport of intensity equation and a self-supervised neural network based on explicit neural fields. Compared with state-of-the-art implicit neural fields, this approach improves computational speed, reconstruction accuracy, and interpretability. We demonstrate FDT in two experimental regimes: reconstructing the 3D RI of label-free tissues up to 300 μ m thick placed on a fluorescent substrate, and reconstructing the 3D RI of fluorescence-labeled, 3D-cultured cells without transmission-side access using fluorescence emitted by the cells themselves. FDT establishes a foundation for accessible reflection-mode multimodal fluorescence-phase imaging in complex biological systems.