High-fidelity reconstruction in structured illumination microscopy through accurate noise modeling and multi-scale wavelet denoising

Structured illumination microscopy (SIM) enables super-resolution imaging in living cells, yet its quantitative accuracy is often compromised by artifacts from linear SIM reconstruction, which amplifies and distorts noise. This signal-dependent noise profile challenges conventional denoising methods, typically resulting in a trade-off between faint signal loss and residual noise retention. To address this, we introduce SURE-SIM, a denoising framework that integrates a multi-scale wavelet transform with Stein’s unbiased risk estimator (SURE), explicitly building upon the Laplace noise model we identified in the SIM reconstruction process. Evaluated on calibration slides and live-cell samples, SURE-SIM improves the peak signal-to-noise ratio (PSNR) by up to 3.1 dB and reduces intensity-correlation errors by over 15%, demonstrating high intensity fidelity in biological imaging. We further validated its performance in a quantitative Förster resonance energy transfer (FRET) workflow that is highly sensitive to intensity distribution. Compared to Wiener-SIM, SURE-SIM reduced the standard deviation of FRET efficiency measurements. This enhancement enabled the clear visualization of FRET signals on mitochondrial structures and significantly reduced quantification uncertainty arising from noise fluctuations.