Fig. 1. Framework of deep learning in super-resolution imaging^[15-27]

Fig. 2. Schematic diagram of STED^[15,16,28]

Fig. 3. Different network framework diagrams and experimental results. (a) MPRNet network architecture^[17]; (b) MPRNet reconstruction images of β-tubulin (STAR635P) in U2OS cells^[17]; (c) STED-flimGANE network structure diagram^[35]; (d) intensity images under extreme conditions of rate depletion^[35]; (e) comparison of nuclear hole imaging between low noise SRDAN and other methods^[18]; (f) line intensity distribution of core hole images in each algorithm^[18]

Fig. 4. Principle of single molecule localization microscopy^[19]

Fig. 5. Network and reconstruction results related to improving image reconstruction speed. (a) Deep-STORM reconstruction effect comparison^[20]; (b) self-STORM network architecture^[42]; (c) self-STORM rebuild effect comparison diagram^[42]; (d) DRSN-STORM Network architecture,feature extracting module (FEM), inference module (IM), and reconstruction module (RM)^[43]; (e) DRSN-STORM microtubule image reconstruction effect^[43]; (f) DECODE reconstruct effect image^[46]

Fig. 6. Reduce the network architecture and reconstruction results related to the frame rate of image reconstruction. (a) ANNA-PAM network architecture^[47]; (b) ANNA-PAM reconstruction compared with PALM^[47]; (c) multi-color single molecule image reconstruction CNN framework^[21]; (d) multicolor single molecule network reconstruction effect^[21]; (e) DBlink network architecture^[49]

Fig. 7. High precision molecular localization related networks and reconstruction results. (a) BGNet network architecture^[22]; (b) background estimation of three PSF imaging methods using BGNet^[22]; (c) reconstruction results of IEEE ISBI microtubule data set are compared, the top half is binary image, and the bottom half is standardized image^[50]; (d) physical simulation feedback flow^[51]; (e) tetrapod and learned PSF localization results^[51]

Fig. 8. Deep learning extracts additional spectral information from PSF networks and imaging results. (a) smNet network architecture and reconstructed image^[52]; (b) color separation and axial positioning architecture and reconstruction effects^[23]; (c) image color classification process diagram of optimized phase mask^[53]; (d) fluorescent-labeled HeLa cell imaging^[53]; (e) color classification image of COS-7 cells^[57]

Fig. 9. SIM imaging optical path and schematic diagram. (a) SIM imaging optical path diagram^[58]; (b) spectrum extension of SIM^[58]; (c) SIM imaging image^[58-60]

Fig. 10. Imaging results of deep learning in structured light super-resolution microscopy imaging. (a) U-Net-SIM3 network architecture^[24]; (b) reconstruction results of U-Net-SIM5 and scU Net under low light conditions^[24]; (c) SIM nanobead imaging using CycleGAN network^[63]; (d) caGAN microtubule imaging under low signal-to-noise ratio^[64]; (e) schematic diagram of the Fourier attention mechanism principle of DFCAN^[65]; (f) DFCAN reconstruction of f-actin cytoskeleton images^[65]

Fig. 11. Network architecture and reconstruction result graph for improving the quality of SIM image reconstruction. (a) RED-fairSIM deep learning network architecture diagram^[25]; (b) RED-fairSIM imaging results of U2OS osteosarcoma cells^[25]; (c) CR-SIM huFIB cell microtubule imaging image^[70]; (d) Deep-MSIM microtubule imaging effect^[72]

Fig. 12. BioSR data set and PINN network related images. (a) BioSR data set^[65]; (b) PINN network architecture^[26]; (c) optimization results of nonlinear SIM resolution based on PINN for multiple object types^[26]

Fig. 13. Deep learning implements cross modal transformation network structure diagram and result diagram. (a) Deep learning implementation of cross modal transformation network framework for fluorescence microscopy^[74]; (b) results of cross modal image conversion from confocal to STED^[74]; (c) partial network structure diagram of TCAN network generator^[27]; (d) DFCAN mechanism network diagram^[27]; (e) TCAN super-resolution imaging results of microtubules^[27]