Fig. 2: Resolution enhancement of VsLFM.

a, MIPs and enlarged regions from xy slices at z = 1 μm of a fixed L929 cell with membrane labeling (TSPAN4-mCherry), obtained by LFM, VCD-Net, HyLFM-Net, VsLFM and sLFM, respectively. b, Boxplots of averaged lateral resolution and axial resolution of LFM, VCD-Net, HyLFM-Net, VsLFM and sLFM at different axial positions (n = 10 beads per plane). The resolution was estimated by imaging 100-nm-diameter fluorescent beads that were uniformly distributed in low-melt agarose with a ×63/1.4 NA oil-immersion objective, and measuring the FWHM with a Gaussian fit. Lateral and axial diffraction-limited resolutions at a center wavelength of 525 nm are shown with the dashed lines for comparison. Data are presented as mean ± s.d. c, Spatial–angular views and the corresponding reconstructed MIPs of a selected 100 nm fluorescent bead, obtained by traditional LFM, VsLFM, sLFM, VCD-Net and HyLFM-Net, respectively. The normalized profiles along the marked dashed lines are shown in the insets. All of the learning-based methods were trained on the bead dataset. d, MIPs of two virtually separated beads obtained by LFM, VCD-Net, HyLFM-Net, VsLFM and sLFM with cross-section profiles along the dashed lines. We imaged the same 100 nm bead at two positions with an interval of 230 nm shifted by a piezo translation stage and added the images together to create the two virtually separated beads. Scale bars, 10 μm (a), 1 μm (c), 200 nm (d).