Skip to main content
. 2020 Apr 22;11:1934. doi: 10.1038/s41467-020-15784-x

Fig. 1. Super-resolution imaging with U-Net.

Fig. 1

a Fifteen or three SIM raw data images were used as input and the corresponding SIM reconstructions from 15 images were used as the ground truth to train the U-Net. Θ: the angle of the sinusoidal patterned illumination; ψ: the phase of the patterned illumination. b Reconstruction results for different subcellular structures. Shown are average projections of 15 SIM raw data images (first column), the reconstruction results from a conventional SIM reconstruction algorithm (second column), U-Net-SIM15 output (third column), U-Net-SIM3 output (fourth column) and line profiles along the dashed line in each image (fifth column). In the line profile plot, the average is shown on the right y-axis and all others share the left y-axis. r indicates the resolution. Shown are representative images randomly selected form the testing dataset indicated in Supplementary Table 1. The training datasets were collected from at least three independent experiments. c The achieved resolution of different approaches was estimated (Source data are provided as a Source Data file). MT microtubules (n = 204); Adh. adhesions (n = 32); Mito. mitochondria (n = 61); Act. F-actin (n = 85). A average; S SIM reconstruction; U15 U-Net-SIM15; U3 U-Net-SIM3. Tukey box-and-whisker plot shown with outliers displayed as dots (Methods). Scale bar: 1 μm.