Fluorescent diffraction tomography using explicit neural fields

Neural fields enable powerful, flexible computational imaging. Optical diffraction tomography (ODT) reconstructs three-dimensional (3D) refractive index (RI) from transmitted light but is typically restricted to thin samples. Fluorescence labeling, however, allows reflection-mode imaging of thicker specimens. We introduce fluorescence diffraction tomography (FDT), which leverages explicit neural fields to reconstruct 3D RI from reflection-mode fluorescence images. Key elements include a coarse-to-fine structure, self-calibration for illumination accuracy, a differentiable multi-slice model, and partially coherent masks for realistic imaging. We demonstrate FDT by reconstructing label-free bovine myotubes in a 530 × 530 × 300 μm³ volume with high resolution. FDT thus extends diffraction tomography to bulky, heterogeneous biological samples.