Fig. 4: Generalization of ZS-DeconvNet to multiple imaging modalities.
From: Zero-shot learning enables instant denoising and super-resolution in optical fluorescence microscopy
a, b Representative confocal (top left), sparse deconvolution (bottom left), and 3D ZS-DeconvNet enhanced (right) images of an early mouse embryo immunostained for microtubule (cyan), chromosomes (orange), actin rings (magenta), and apical ___domain (green). c, d Magnified regions of microtubule bridges (c) and actin rings (d) labeled with white dashed boxes in (a) and (b) acquired via confocal microscopy, sparse deconvolution, and 3D ZS-DeconvNet. e Representative WF (center region) and 3D ZS-DeconvNet enhanced (surrounding region) images of a C. elegans embryo with apical junction, cell membrane (cyan) and lysosomes (red) labeled. f, g Lysosome channel of the central region in (e) color-coded for distance from the substrate. Both WF (f) and 3D ZS-DeconvNet processed images (g) are shown for comparison. h Time-lapse 3D ZS-DeconvNet enhanced images showing the process of hypodermal cell fusion (red arrows) during the development of a C. elegans embryo. Scale bar, 5 μm (a, b, e), 2 μm (c, d), 3 μm (g, h), 1 μm (zoom-in region of g). Gamma value, 0.7 for cytomembrane and lysosomes in the C. elegans embryo.
