Adaptive multiband Dark sectioning for robust background removal in fluorescence microscopy

Abstract

Out-of-focus fluorescence and scattering background degrade image contrast and compromise quantitative analysis in fluorescence microscopy, particularly in thick biological specimens. Conventional Dark sectioning suppresses diffuse background through low-frequency dark-channel correction, but its binary frequency partition and globally defined parameters can over-suppress broad fluorescent structures whose genuine low-frequency content overlaps the background, producing hollow artifacts and depth-dependent structural loss. Here, we propose Adaptive multiband Dark sectioning (Adaptive Dark), which replaces the binary split with Fourier-domain multiband decomposition and performs subband-specific correction. Foreground thresholds and suppression strengths are estimated automatically for the selected low-frequency subbands, whereas intermediate- and high-frequency components are preserved and the corrected low-frequency components are weighted during reconstruction. Ground-truth simulations, ablation experiments, and data from wide-field, confocal, two-photon, structured illumination, and stimulated emission depletion microscopy demonstrate improved structural preservation while maintaining effective background suppression. In simulation, Adaptive Dark achieved a structural similarity index (SSIM) of 0.83 versus 0.24 for Conventional Dark, while the corresponding signal-to-background ratio (SBR) values were 18.4 and 17.3, respectively. Deep two-photon datasets further showed higher SSIM and Pearson correlation coefficient with effective SBR enhancement across imaging depth. Adaptive Dark requires no hardware modification or training data and reduces dependence on dataset-specific manual tuning, providing a practical post-acquisition approach for robust fluorescence-background removal.

The authors' abstract, as published at the source. Biomedical Optics Express, 2026 · DOI ↗