Physics-driven self-supervised learning for quantitative high-fidelity structured illumination microscopy

Abstract

Structured illumination microscopy (SIM) enables rapid, long-term super-resolution (SR) imaging of live-cell dynamics. However, current SIM reconstruction methods, although achieving high structural fidelity, generally lack reliable intensity quantification. Here we present qHiFi-SIM, a physics-driven self-supervised learning framework for quantitative high-fidelity SIM imaging. By leveraging the wide-field (WF) image as a physical intensity reference and conventional SIM reconstruction as a structural continuity prior, qHiFi-SIM enables self-supervised training without ground-truth intensity data. It achieves structural fidelity comparable to HiFi-SIM while maintaining excellent intensity linearity at the SIM point-spread function scale (R2 > 0.986, mean intensity ratio near 1). The framework transfers across diverse SIM setups and typical samples, and directly improves the intensity linearity of other state-of-the-art algorithms without retraining. We demonstrate its utility for live-cell quantitative visualization of mitochondrial intensity dynamics and high-fidelity SIM-FRET functional imaging, offering a practical tool for SR structural visualization and quantitative functional imaging in live cells.