Diffusion-Based Translation of Densely Labeled Actin into Fluorescent Speckle Microscopy Data

Understanding cytoskeletal function requires measuring both actin architecture and its remodeling over time, yet these complementary properties are typically captured using different imaging strategies. Densely labeled actin reveals the organization of the cytoskeletal network, whereas fluorescence speckle microscopy (FSM) enables quantitative measurement of actin movement and turnover but provides only a sparse view of its architecture. We investigated whether a diffusion-based image-to-image translation model could computationally connect these representations by converting continuous actin images into speckle-like images suitable for downstream analysis. Baseline CycleGAN-Turbo outputs contained recognizable speckles but were elongated and overly dense. We therefore introduced two loss terms, one based on the microscope’s point-spread-function and the other based on signal sparsity to favor compact, isolated fluorescent features. The modified model generated PSF shaped, speckles while preserving spatial correspondence with the input actin architecture. In paired time-lapse data, speckle displacement measured from synthetic movies distinguished unperturbed cells from paraformaldehyde (PFA)-treated cells. These results demonstrate the feasibility of deriving an FSM-compatible representation from densely labeled actin images, providing a step toward obtaining complementary information about actin architecture and dynamics from a single microscopy acquisition. Further development will focus on preserving temporal fidelity for quantitative analysis of actin remodeling.