Parameter-free clear image deconvolution (CLID) technique for single-frame live-cell super-resolution imaging

Abstract

Advancing single-frame imaging techniques beyond the diffraction limit and upgrading traditional wide-field or confocal microscopes to super-resolution (SR) capabilities are greatly sought after by biologists. While enhancing image resolution by deconvolving noise-free images is beneficial, achieving a noise-free image that maintains the distribution of signal intensity poses a challenge. We first developed a denoising method utilizing reversibly switchable fluorescent proteins through synchronized signal switching (3S). Additionally, we introduced a denoising neural network technique, 3Snet, which combines supervised and self-supervised learning using 3S denoised images as the ground truth. These approaches effectively eliminate noise while maintaining fluorescence signal distribution across camera pixels. We then implemented clear image deconvolution (CLID) on both 3S and 3Snet denoised images to develop SR techniques, named 3S-CLID and 3Snet-CLID. Notably, 3Snet-CLID boosts the resolution of single fluorescence images from wide-field and spinning-disk confocal microscopies by up to 3.9 times, achieving a spatial resolution of 65 nm, the highest in such imaging scenarios without an additional SR module and complex parameter setting. 3Snet-CLID enables dual-color single-frame live-cell imaging of various subcellular structures labeled with conventional fluorescent proteins and/or dyes, allowing observations of dynamic cellular processes. We expect that these advancements will drive innovation and uncover new insights in biology.