Independent noise realizations enable morphologically agnostic image reconstruction in single-photon-sensitive microscopy

The study introduces a novel regularisation framework for Richardson-Lucy (RL) deconvolution in single-photon sensitive fluorescence microscopy. This framework, named Regularized by Noise (RbN), relies on independent noise realisations rather than object morphology assumptions. By preserving Poisson statistics, RbN distinguishes actual image features from stochastic noise, selecting the optimal regularisation strength through the Poisson residual whiteness principle.

This data-driven approach eliminates the need for heuristic parameter tuning and prevents the semi-convergent behavior typically associated with RL deconvolution. The framework was experimentally validated on photon-timing-resolved confocal and image scanning microscopy across eight distinct subcellular targets, demonstrating its effectiveness across various imaging modalities, detector technologies, biological structures, and signal-to-noise regimes.

The method successfully overcomes RL's semi-convergent behavior, removes dependence on stopping criteria, and consistently outperforms conventional RL while preserving fine structural details. The authors propose that independent noise realisations can serve as a general source of morphology-agnostic regularisation, potentially extending to other imaging modalities and statistical inverse problems with suitable noise-specific formulations.