Extended Data Fig. 17: Multi-modality imaging enabled with the multiview line confocal system.

a) Different methods of combining data, enabling a highly versatile imaging platform. Left: Diffraction limited volumes acquired from views A (yellow), B (green), C (red) may be combined with joint deconvolution to yield triple-view diffraction-limited data (RYG arrows). Middle: Alternatively, 5 volumes per view may be collected and processed as in Extended Data Fig. 10 for 1D SIM, and the 3 1D SIM volumes combined using joint deconvolution to reconstruct triple-view 1D SIM data. Right: Instead, confocal data from each view may be passed through 1D SIM networks, and the data combined via joint deconvolution. Combining 6 rotations (for clarity only two are shown in figure) from view C yields view C 2D SIM data. If the procedure is repeated for views A, B, and the data combined with joint deconvolution, triple-view 2D SIM data may be obtained. xz cross-sections through PSFs are shown in black boxes. Blue volumes are relative sizes of PSFs that result from each process. Scale bars: white: 500 nm, black: 200 nm. b) Applications include wide-field microscopy, single-view line confocal microscopy (from any of the views), single-view 1D SIM, triple-view diffraction limited imaging, and triple-view 1D SIM. With deep learning (red), triple-view line confocal volumes can be predicted from low SNR single-view input, 1D SIM can be predicted from diffraction-limited input, and combination with joint deconvolution allows further extension to single- and triple-view 2D SIM. Biological and imaging performance examples (resolution, imaging speed and duration) are also provided. Resolution values in line confocal microscopy, triple-view line confocal (without deep learning), and triple-view 1D-SIM (without deep learning) are estimated from immunostained microtubules in fixed U2OS cells. Deep learning resolution values are estimated from fine C. elegans embryo neurites (triple-view confocal) or actin fibres (single view 2D SIM, triple-view 2D SIM). See also Supplementary Table 1. c) Decorrelation resolution analysis from the images of worm L4 larval (strain DCR8528) expressing membrane targeted GFP primarily in the nervous system. See also Extended Data Fig. 18d and Fig. 4i–m. Data (mean ± standard deviation) are derived from 45 measurements (3 animals, 15 planes per animal). Given the ~2.3-fold improvement laterally and ~2.6-fold improvement axially, the triple-view 2D SIM (253 × 253 × 322 nm³) result offers a volume resolution improvement of ~13.8-fold over the raw view C data (601 × 56 × 836 nm³). d) Apparent widths (open circles, mean ± standard deviation) of 8 actin fibres from the cell presented in Fig. 4n–q, comparing lateral (left) and axial (right) full width at half maximum in different microscope modalities. Given the ~2.2-fold improvement laterally and ~2.4-fold improvement axially, the triple-view 2D SIM result offers a volume resolution improvement of ~11.6-fold over the raw view C data.