ROOF: Robustness-Oriented Optimal Factorization for Fluorescence Crosstalk Correction

MOTIVATION: Fluorescence bleed-through (spectral crosstalk) compromises the accuracy of signal interpretation, quantification, and visualization in multi-band fluorescence microscopy. Existing correction methods rely either on experimental calibration or blind spectral unmixing, but often suffer from limited robustness to biological variability, overcorrection artifacts, or the need for manual parameter tuning. We propose ROOF (Robustness-Oriented Optimal Factorization), a semi-blind framework that unifies calibration-based modeling with data-driven spectral unmixing. It formulates crosstalk correction as a diagonally normalized nonnegative matrix factorization problem regularized by a reference crosstalk matrix obtained from single-labeled control samples. The regularization parameter is estimated automatically via maximum likelihood, eliminating manual tuning while adapting to specimen-specific variations. RESULTS: Experiments on synthetic and real four-band fluorescence images demonstrate that ROOF consistently outperforms state-of-the-art crosstalk correction methods. ROOF achieves higher correction accuracy, suppresses overcorrection artifacts, preserves biologically meaningful fluorescence colocalization, and is computationally efficient. These results highlight ROOF's robustness, efficiency, and practical value for reliable quantitative fluorescence imaging. AVAILABILITY AND IMPLEMENTATION: The codes are available at https://github.com/baobabking/ROOF with DOI:10.5281/zenodo.21872937.