Non-invasive fluorescence imaging through scattering media via variance-guided wavefront shaping and convex optimization-based deconvolution

Scattering media pose a significant barrier to non-invasive biomedical imaging, as conventional wavefront shaping methods rely on invasive guide stars or costly nonlinear modalities. Here, we introduce an improved approach that enables high-fidelity, non-invasive fluorescence imaging through scattering media by combining the linear fluorescence mechanism with efficient computational optimization. The method leverages a genetic algorithm guided by variance maximization to dynamically optimize speckle, non-invasively exciting an individual fluorescent bead by ∼10-fold enhancement in target intensity ratio. This process generates a precise system point spread function (PSF), which drives a convex optimization-based deconvolution framework to reconstruct obscured targets. Remarkably, the technique eliminates the need for complex scanning systems, achieving rapid wide-field imaging with structural similarity (SSIM) indices exceeding 0.997 (for beads). We demonstrate robust imaging of both discrete beads and continuous fibers behind scattering media, revealing resolution superior to that of conventional speckle cross-correlation methods. The method provides a pathway for non-invasively visualizing fluorescent objects behind scattering media.