Fig. 1. Development of super-resolution photoluminescence imaging techniques^[10]

Fig. 2. Schematic diagrams of the principle and experimental setup of MINFLUX. (a) (b) Schematic diagrams of the principle to localize the center of fluorescent molecules for common nanoscopy and MINFLUX, respectively^[17]; (c) schematic diagram of the setup of MINFLUX^[18]; (d)-(f) principles of MINFLUX illustrated in one, two, and three dimensions, respectively^[16,18]; (g) comparison of the spatial resolution between SMLM and MINFLUX under same photon counts (scale is 10 nm)^[16]

Fig. 3. Development in terms of resolution and 3D multicolor imaging of MINFLUX. (a) Two-color MINFLUX nanoscopy in 3D of U-2 OS cell with AF647 and CF680(scale: 500 nm)^[19]; (b) three-dimensional localization accuracy of MINFLUX^[20]; (c) schematic of the p-MINFLUX setup^[21]; (d)(e) 2D localization image and fluorescence lifetime image for one DNA origami structure through p-MINFLUX^[21]; (f)(g) measurement method based on grating scanning single molecule localization and imaging simulation results of this method at different scanning distances^[23]

Fig. 4. Super-resolution imaging by combining the MINFLUX with other techniques. Simulation of z-axis localization for (a) single-photon MINFLUX and (b) two-photon MINFLUX, respectively(scale: 1 nm)^[25];(c) concept of ISM-FLUX, an activation laser beam activates a single fluorophore in the sample (yellow star) which is sequentially excited by a series of spatially displaced doughnut beams^[27]; (d) SMLM and (e) SIMFLUX image for the same nano-rulers(scale: 50 nm)^[28]; (f) comparison of imaging effects of confocal microscopy, (g) stimulated radiation loss microscopy, and (h) minimum photon number stimulated radiation loss technique on similar mitochondria (scale: 200 nm)^[29]; (i) comparison of confocal fluorescence imaging using Vimentin-reEGFP2 fluorescent protein (left image) and minimum photon number microscopy imaging using DNA-PAINT technology (scale: 200 nm)^[31]

Fig. 5. Comparison of wide field imaging, minimum photon number microscopic imaging technology, and their combination with other super-resolution methods in field of view, photon efficiency, signal to back ratio, system simplicity, and resolution ability

Fig. 6. Principle of SOFI^[32]. (a) Emitter distribution in the object plane; (b) magnified detail of the dotted box in Fig. (a); (c) time trajectory of each pixel; (d) second-order correlation function is calculated from the fluctuations for each pixel; (e) result of integrating the second-order correlation function per pixel

Fig. 7. SOFI images with different fluorescent labels. (a) Schematic diagram of joint labeled wave super-resolution microscopy imaging^[37]; (b) schematic cross-section of the joint-tagging SOFI imaging^[37]; (c) workflow of multicolor SOFI imaging by spectral cross-cumulant analysis followed by linear unmixing using simulations^[38]; (d) conventional and (e) pcSOFI image of HeLa cell labeled with Lyn-Dronpa (green) and Kras-rsTagRFP (red), respectively^[41] (scale: 10 μm)

Fig. 8. Flowchart of algorithm to solve artifacts in the high-order SOFI. (a) Flowchart of local dynamic range compression algorithm^[44]; (b) flowchart to illustrate the different steps of the bSOFI algorithm^[45]

Fig. 9. Flowchart of the analysis of SOFI data and the contrast between SOFI and other imaging techniques. (a) Schematic overview of a super-resolution localization or SOFI analysis^[46]; (b) comparison of the spatial distribution of individual chromophores and their wide field imaging, balanced super-resolution wave microscopy imaging, and random optical reconstruction super-resolution imaging^[48]; (c) comparison of confocal fluorescence microscopy, image scanning microscopy, and second-order and fourth-order super-resolution optical wave image scanning microscopy for commercial quantum dot QD625^[51]

Fig. 10. Principle of super-resolution image based on anti-bunching effect. (a) Schematic diagram of the HBT measurement^[61]; (b) anti-bunching signal of single molecule excited by continuous laser^[62]; (c)-(e) first order, second order, and third order anti-bunching images of quantum dots in the same region^[64]

Fig. 11. Imaging results of super-resolution image based on anti-bunching effect. (a) Schematic diagram of a super-resolution imaging device for measuring second-order anti-bunching in a confocal system^[65]; (b) reconstructed image according to the single and double photon signals of NVs collected from the system in a^[65]; (c) schematic diagram of a super-resolution imaging device for measuring high-order anti-bunching with a single-photon fiber beam camera^[67]; (d) two-dimensional localization accuracy measured for a single QD (solid blue) and theoretical accuracy by using the system in c^[67]; (e) schematic of SIQCM by combining SIM and quantum correlation microscopy^[68]; (f) simulation results illustrated the resolution of SIQCM^[68]; (g) (h) imaging results of QD625-labeled microtubule cell samples in confocal microscopy and Q-ISM^[69]

Fig. 12. Development of neural network architecture. (a) Taxonomy of AI^[72]; (b) overall architecture of the CNN includes an input layer, multiple alternating convolution, and max-pooling layers, one fully-connected layer, and one classification layer^[72]; (c) GAN model frame diagram^[75]; (d) U-net architecture^[76]

Fig. 13. Latest progress in fluorescence super-resolution microscopy imaging based on deep learning. (a) Reconstruction results of wide-field images of bovine pulmonary artery endothelial cells through deep learning^[78]; (b) total internal reflection fluorescence microscopic effect based on depth learning, and its comparison with structured light super-resolution microscopic imaging based on total internal reflection, the cell used is SUM159^[78]; (c) resolution characterization of GAN net ^[79]; (d) based on unsupervised content retention transformation, microscopic imaging can convert wide-field images into super-resolution images, enabling the resolution of sub diffractive structures such as microtubules and secretory particles from the wide-field images^[80]; (e) statistical comparison of normalized root mean square error, multi-scale structure similarity index, and resolution of 121 groups of actin images reconstructed by scU-net, DFCAN, and DFGAN networks respectively, black cross is the outlier^[82]

Fig. 14. Deep learning-enabled transformation of images of curves from 1d_SIM to 9_SIM^[83]. (a) WF curve image; (b) 1d_SIM image is network input; (c) 3_SIM image is network output; (d) 9_SIM image is real images. Reconstruction effect of microtubule imaging based on U-Net network^[81]. (e) Average projection of 15 SIM raw data images; (f) reconstruction results of reconstruction algorithm of traditional structured light illumination microscopic imaging; (g) output effect of U-Net-SIM15 network; (h) output effect of U-Net-SIM3 network. Comparison of results of ANNA-PALM reconstruction of immunostaining microtubule microscopy images^[85]. (i) Widefield image; (j) sparse PALM image obtained from the first 9 s of acquisition (k=300 frame, n=11740 localizations); (k) dense PALM image obtained from a 15 min-long acquisition (K=30000 frame, N=409364 localizations); (l) ANNA-PALM reconstruction from the widefield Fig. (i) only; (m) ANNA-PALM reconstruction from the sparse PALM Fig. (j) only; (n) ANNA-PALM reconstruction from the widefield Fig. (i) and sparse PALM Fig. (j) combined (scale: 1 µm). Two-color sSMLM images in COS-7 cell^[86]. (o) Low-density image with 3000 frame; (p) deep CNN reconstruction; (q) high-density image with 19997 frame (pixel size:16 nm. scale: 1.5 µm)

Table 1. Parameters of the deep-learning super-resolution imaging