Fig. 1. Schematic diagrams of the imaging optical paths and characteristics^[16].(a) Confocal fluorescence microscopy; (b) light sheet fluorescence microscopy; (c) light field fluorescence microscopy

Fig. 2. (a)‒(d) Deep learning enables light sheet fluorescence microscopy to break through the diffraction limit and improving the spatial resolution of imaging, the spatial resolution and time series results of DR-SPIM (uncombined deep learning method) and IDDR-SPIM (combined deep learning method) are displayed and compared^[29]; (e) (f) deep learning enables structured light microscopy to improve imaging temporal resolution, the figures contain schematic diagrams and reconstruction results for microtubules^[30]; (g)‒(j) deep learning enables single-molecule localization microscopy to improve imaging temporal resolution, the figures contain schematic diagrams and comparisons with other methods^[34]

Fig. 3. (a) VCD-Net reconstruction process;(b)(c) MIP of one instantaneous volume of xy (top) and xz (bottom) sides of the beating cardiomyocyte nucleus calculated by VCD-LFM and LFDM, respectively; (d)(e) visualized in 3D by VCD-LFM and LFDM at a time point, respectively; (f) myocardial volume during diastolic and systolic phases in a cardiac cycle; (g) the rate of volume change in central ventricular diastole and systolic in a cardiac cycle^[38]

Fig. 4. (a)‒(e) Schematic diagrams of the process and effect of CACS which enables light sheet fluorescence microscopy to improve imaging resolution^[42]; (h)‒(i) Self-Net enables fluorescence microscopy to improve axial resolution, comparison of raw data and self-net output data^[43]