Abstract
Deep-learning-based structured illumination microscopy (SIM) has demonstrated substantial potential in long-term super-resolution imaging of biostructures, enabling the study of subcellular dynamics and interactions in live cells. However, the acquisition of ground-truth (GT) data for training poses inherent challenges, limiting its universal applicability. Current approaches without using GT training data compromise reconstruction fidelity and resolution, and the lack of physical priors in end-to-end networks further limits these qualities. Here we developed self-supervised reconstruction (SSR)-SIM by combining statistical analysis of reconstruction artifacts with structured light modulation priors to eliminate the need for GT and improve reconstruction precision. We validated SSR-SIM on common biological datasets and demonstrated that SSR-SIM enabled long-term recording of dynamic events, including cytoskeletal remodeling in cell adhesion, mitochondrial cristae remodeling, interactions between viral glycoprotein and endoplasmic reticulum, endocytic recycling of transferrin receptors, vaccinia-virus-induced actin comet remodeling, and mitochondrial intercellular transfer through tunneling nanotubes.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$32.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 print issues and online access
$259.00 per year
only $21.58 per issue
Buy this article
- Purchase on SpringerLink
- Instant access to the full article PDF.
USD 39.95
Prices may be subject to local taxes which are calculated during checkout
Similar content being viewed by others
Data availability
The published BioSR dataset9 was used to validate SSR-SIM and other SIM algorithms (Fig. 1f and Extended Data Figs. 2, 4 and 5). Due to the substantial size of long-term live-cell imaging data, depositing the complete dataset in a public repository is not currently feasible. However, representative F-actin imaging datasets for training and testing SSR-SIM models are publicly available via figshare at https://doi.org/10.6084/m9.figshare.30406366.v1 (ref. 48). Any additional live-cell imaging datasets are available from the corresponding authors upon request. Source data are provided with this paper.
Code availability
The SSR-SIM Python code is available on GitHub at https://github.com/HUST-Tan/SSR-SIM (MIT license). Example training and testing datasets and pretrained models are available via figshare at https://doi.org/10.6084/m9.figshare.30406366.v1 (ref. 48).
References
Betzig, E. et al. Imaging intracellular fluorescent proteins at nanometer resolution. Science 313, 1642–1645 (2006).
Gustafsson, M. G. L. Surpassing the lateral resolution limit by a factor of two using structured illumination microscopy. J. Microsc. 198, 82–87 (2000).
Hell, S. W. & Wichmann, J. Breaking the diffraction resolution limit by stimulated emission: stimulated-emission-depletion fluorescence microscopy. Opt. Lett. 19, 780–782 (1994).
Rust, M. J., Bates, M. & Zhuang, X. Sub-diffraction-limit imaging by stochastic optical reconstruction microscopy (STORM). Nat. Methods 3, 793–796 (2006).
Li, D. et al. Extended-resolution structured illumination imaging of endocytic and cytoskeletal dynamics. Science 349, aab3500 (2015).
Chu, K. et al. Image reconstruction for structured-illumination microscopy with low signal level. Opt. Express 22, 8687–8702 (2014).
Huang, X. et al. Fast, long-term, super-resolution imaging with Hessian structured illumination microscopy. Nat. Biotechnol. 36, 451–459 (2018).
Jin, L. et al. Deep learning enables structured illumination microscopy with low light levels and enhanced speed. Nat. Commun. 11, 1934 (2020).
Qiao, C. et al. Evaluation and development of deep neural networks for image super-resolution in optical microscopy. Nat. Methods 18, 194–202 (2021).
Qiao, C. et al. Rationalized deep learning super-resolution microscopy for sustained live imaging of rapid subcellular processes. Nat. Biotechnol. 41, 367–377 (2023).
Chen, J. et al. Three-dimensional residual channel attention networks denoise and sharpen fluorescence microscopy image volumes. Nat. Methods 18, 678–687 (2021).
Ward, E. N. et al. Machine learning assisted interferometric structured illumination microscopy for dynamic biological imaging. Nat. Commun. 13, 7836 (2022).
Lehtinen, J. et al. Noise2Noise: learning image restoration without clean data. In Proc. International Conference on Machine Learning (eds Dy, J. et al.) 2965–2974 (PMLR, 2018).
Huang, T., Li, S., Jia, X., Lu, H. & Liu, J. Neighbor2neighbor: self-supervised denoising from single noisy images. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (eds Masson-Forsythe, M. et al.) 14776–14785 (IEEE, 2021).
Chen, X. et al. Self-supervised denoising for multimodal structured illumination microscopy enables long-term super-resolution live-cell imaging. PhotoniX 5, 4 (2024).
Li, X. et al. Reinforcing neuron extraction and spike inference in calcium imaging using deep self-supervised denoising. Nat. Methods 18, 1395–1400 (2021).
Gustafsson, M. G. et al. Three-dimensional resolution doubling in wide-field fluorescence microscopy by structured illumination. Biophysical J. 94, 4957–4970 (2008).
McCann, M. T., Jin, K. H. & Unser, M. Convolutional neural networks for inverse problems in imaging: a review. IEEE Signal Process. Mag. 34, 85–95 (2017).
Kong, F. et al. Residual local feature network for efficient super-resolution. In Proc. IEEE Conference on Computer Vision and Pattern recognition Workshops (eds Mortensen, E. et al.) 766–776 (IEEE, 2022).
Zamir, S. W. et al. Restormer: efficient transformer for high-resolution image restoration. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (eds Mortensen, E. et al.) 5728–5739 (IEEE, 2022).
Dosovitskiy, A. et al. An image is worth 16×16 words: transformers for image recognition at scale. In Proc. International Conference on Learning Representations (eds Mohamed, S. et al.) 611–631 (ICLR, 2021).
Gal, Y. & Ghahramani, Z. Dropout as a Bayesian approximation: representing model uncertainty in deep learning. In Proc. International Conference on Machine Learning (eds Balcan, M. F. et al.) 1050–1059 (PMLR, 2016).
Guo, Y. et al. Visualizing intracellular organelle and cytoskeletal interactions at nanoscale resolution on millisecond timescales. Cell 175, 1430–1442 (2018).
Liu, T. et al. Multi-color live-cell STED nanoscopy of mitochondria with a gentle inner membrane stain. Proc. Natl Acad. Sci. USA 119, e2215799119 (2022).
Kilian, N. et al. Assessing photodamage in live-cell STED microscopy. Nat. Methods 15, 755–756 (2018).
Orrenius, S., Zhivotovsky, B. & Nicotera, P. Regulation of cell death: the calcium–apoptosis link. Nat. Rev. Mol. Cell Biol. 4, 552–565 (2003).
Presley, J. F. et al. ER-to-Golgi transport visualized in living cells. Nature 389, 81–85 (1997).
Guimaraes, S. C. et al. Peroxisomes, lipid droplets, and endoplasmic reticulum ‘hitchhike’ on motile early endosomes. J. Cell Biol. 211, 945–954 (2015).
Grant, B. D. & Donaldson, J. G. Pathways and mechanisms of endocytic recycling. Nat. Rev. Mol. cell Biol. 10, 597–608 (2009).
Lum, F.-M. et al. Monkeypox: disease epidemiology, host immunity and clinical interventions. Nat. Rev. Immunol. 22, 597–613 (2022).
Humphries, A. C. et al. Clathrin potentiates vaccinia-induced actin polymerization to facilitate viral spread. Cell Host Microbe 12, 346–359 (2012).
Horsington, J., Turnbull, L., Whitchurch, C. B. & Newsome, T. P. Sub-viral imaging of vaccinia virus using super-resolution microscopy. J. Virol. Methods 186, 132–136 (2012).
Rustom, A., Saffrich, R., Markovic, I., Walther, P. & Gerdes, H.-H. Nanotubular highways for intercellular organelle. Transp. Sci. 303, 1007–1010 (2004).
Pinto, G., Brou, C. & Zurzolo, C. Tunneling nanotubes: the fuel of tumor progression? Trends Cancer 6, 874–888 (2020).
York, A. G. et al. Resolution doubling in live, multicellular organisms via multifocal structured illumination microscopy. Nat. Methods 9, 749–754 (2012).
Chen, B.-C. et al. Lattice light-sheet microscopy: imaging molecules to embryos at high spatiotemporal resolution. Science 346, 1257998 (2014).
Müller, C. B. & Enderlein, J. Image scanning microscopy. Phys. Rev. Lett. 104, 198101 (2010).
Fiolka, R., Beck, M. & Stemmer, A. Structured illumination in total internal reflection fluorescence microscopy using a spatial light modulator. Opt. Lett. 33, 1629–1631 (2008).
Chen, L. et al. Sca-cnn: spatial and channel-wise attention in convolutional networks for image captioning. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (eds Mortensen, E. et al.) 5659–5667 (IEEE, 2017).
Vaswani, A. et al. Attention is all you need. In Proc. International Conference on Neural Information Processing Systems (eds Guyon, I. et al.) 5998–6008 (PMLR, 2017).
Shi, W. et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (eds Mortensen, E. et al.) 1874–1883 (IEEE, 2016).
Kingma, D. P. & Ba, J. Adam: a method for stochastic optimization. In Proc. International Conference on Learning Representations (eds Bengio, Y. et al.) 1–15 (ICLR, 2015).
Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat. Methods 9, 676–682 (2012).
Sternberg, S. R. Biomedical image processing. Computer 16, 22–34 (1983).
Galloway, C. M., Le Ru, E. C. & Etchegoin, P. G. An iterative algorithm for background removal in spectroscopy by wavelet transforms. Appl Spectrosc. 63, 1370–1376 (2009).
Tinevez, J.-Y. et al. TrackMate: an open and extensible platform for single-particle tracking. Methods 115, 80–90 (2017).
Descloux, A., Grußmayer, K. S. & Radenovic, A. Parameter-free image resolution estimation based on decorrelation analysis. Nat. Methods 16, 918–924 (2019).
Liu, J. & Li, D. SSR-SIM dataset. figshare https://doi.org/10.6084/m9.figshare.30406366.v1 (2025).
Acknowledgements
We thank K. He (Institute of Genetics and Developmental Biology, Chinese Academy of Sciences) for providing the VSVG-mEmerald plasmid DNA and the SUM159 cell line expressing Rab11a-GFP (KI). This work was supported by grants from the National Natural Science Foundation of China (NSFC; 92254306 (D.L.), 32125024 (D.L.), 62071197 (S.T.), 62471192 (S.T.)); STI2030-Major Projects (2022ZD0206700 (D.L.)), the National Key R&D Program of China (2024YFA1307400 (D.L.)), Beijing Natural Science Foundation (Z240012 (D.L.)) and the New Cornerstone Science Foundation (D.L.). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
Author information
Authors and Affiliations
Contributions
D.L., S.T. and Y.-J.L. supervised the research; J.L. and H. Liu developed the algorithm under the supervision of D.L., S.T. and E.Y.L.; X.D., X.H. and H. Lu built the hardware systems under the supervision of D.L.; J.L., Q.M. and H. Lu developed the imaging and reconstruction software; W.L., T.J., A.J. and X.G. prepared the samples; J.L., H. Lu, W.L., T.J. and X.D. performed the imaging experiments under the supervision of D.L. and Y.-J.L.; J.L., W.L., T.L. and J.C. analyzed the data and prepared the figures and videos with conceptual advice from D.L. and S.T.; D.L., S.T. and J.L. wrote the paper with input from all authors. All authors participated in discussions and commented on the paper.
Corresponding authors
Ethics declarations
Competing interests
S.T., J.L. and D.L. have a pending patent on the presented framework. The other authors declare no competing interests.
Peer review
Peer review information
Nature Methods thanks Ming Lei and Shian Zhang for their contribution to the peer review of this work. Peer reviewer reports are available. Primary Handling Editors: Rita Strack and Nina Vogt, in collaboration with the Nature Methods team.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Extended Data Fig. 1 Assessments of different SSR sampling strategies for high spatiotemporal resolution imaging.
a,b, Simulation of moving hollow vesicles (diameter: 120 nm) under temporal (a, TI-SSR) and spatial (b, SI-SSR) sampling strategies. The alignment error (AE, defined as 100% minus the cosine similarity between paired sampled noisy images) and generalization error (GE, defined as 100% minus the cosine similarity between a sampled noisy image and its corresponding source image) are shown. c, AE for TI-SSR across varying average displacements per frame (10-50 nm) at a fixed camera sampling rate of 60 nm/pixel. d,e, AE (d) and GE (e) for SI-SSR across varying camera spatial sampling rates (20-80 nm/pixel); no sample displacement was applied. f-i, NRMSE and SSIM metrics of TI-SSR and SI-SSR outputs under deviated self-supervised prerequisites (n = 100). Central line, medians; limits, 75% and 25%; whiskers, maximum and minimum. j,k, Proportion of resolved hollow vesicles under different sampling settings.
Extended Data Fig. 2 Comparison of supervised and self-supervised iterative optimization of PHCT.
a, The noisy raw image of F-actin with spectrum (middle) and decorrelation analysis (right). \({\rm{Kc}}\) denotes the cut-off frequency measured in the ratio of the Nyquist frequency. b, The corresponding ground-truth (GT) image. c,d, Validation image patches and metrics for supervised optimization (c) and self-supervised optimization (d) at iteration numbers 1,000, 2,000, 3,000, 5,000, and 200,000. Scale bar, 2 μm (a,b).
Extended Data Fig. 3 Assessments of the impact of excitation pattern density on deep learning-based SIM.
a, The pattern line spacing varies with the excitation NA. Yellow arrows point to representative regions in the high-excitation-NA reconstruction. Experiments were repeated independently for 5 ROIs with similar results. b, PSNR and SSIM comparisons of PHCT with/without the PPA block at varying excitation NA and pattern line spacing, n = 20. Central line, medians; limits, 75% and 25%; whiskers, maximum and minimum. Scale bar, 1 μm (a).
Extended Data Fig. 4 Ablation study of the PPA block.
Representative SSR-SIM images of microtubules and F-actin using PHCT, with and without the PPA block. Arrows highlight fine microfilament connections and adjacent microtubules. PSNR, SSIM, and NRMSE values are attached for reference. Experiments were repeated independently for 5 ROIs (in different cells) with similar results. APC, average photon count of raw data. Scale bar, 2 μm.
Extended Data Fig. 5 Comparison of SSR-SIM with other SIM algorithms.
Red arrows indicate representative hollow ring structures of CCPs, adjacent microtubules, and dense F-actin. The NRMSE metric for each reconstruction is provided. Experiments were repeated independently for 5 ROIs with similar results. APC, average photon count. Scale bar, 1 μm.
Extended Data Fig. 6 Reconstruction and uncertainty quantification of SSR-SIM models via Monte Carlo Dropout.
a, Schematic of the SSR-SIM architecture. Input/output layers are shown in blue and hidden layers in orange. b, Diagram of the Monte Carlo Dropout (MCD) strategy used for uncertainty quantification. MCD (dropout rate p = 0.05) was applied before all learnable layers during both training and inference. c, Multiple stochastic reconstructions generated from a single input using the trained SSR-SIM model with MCD. d,e, The mean reconstruction (d) and standard deviation (uncertainty, e) computed from the multiple outputs shown in c. Models were trained on microtubule data (MTs, Domain A) and Clathrin-coated pits data (CCPs, Domain B), and tested on microtubule data (Domain A). For each training condition, the final mean reconstruction and the corresponding uncertainty maps are displayed. Experiments were repeated independently for 5 cells with similar results. Scale bars, 10 μm (c) and 3 μm (d).
Extended Data Fig. 7 Assessment of cellular activity and photobleaching under conventional SIM and SSR-SIM imaging.
a, Ca2+-level response in Fluoforte-dyed and Calnexin-Halo-expressing COS-7 cells under negative control condition (Fluoforte excited in 488-nm laser, without 561-nm-channel excitation or SIM illumination) (n = 14). Imaging for each ROI was acquired over 10 minutes (600 time points); same for all subsequent panels. b,c, Ca2+-level response over time during conventional SIM (b) and SSR-SIM (c) live-cell imaging (n = 15 each). d,e, Endoplasmic reticulum (ER) fluorescence signals over time during conventional SIM (d) and SSR-SIM (e) imaging. (n = 15 each). f, Representative Ca2+ signal images under negative control condition. g-j, Representative Ca2+ signal (g,i) and ER images (h,j) before and after conventional SIM and SSR-SIM imaging. Experiments (f-j) were repeated independently for 15 cells with similar results. Scale bars: 15 μm (full-field views in f-j) and 5 μm (zoomed-in regions in g-j).
Extended Data Fig. 8 SSR-SIM imaging of VSVG and sheet-like ER interaction.
a, SSR-SIM imaging of co-movement of VSVG and ER structures. b, Time-lapse images of the dynamic events in cyan-boxed region (a). c, A separate image showing additional VSVG and ER interactions. d, Time-lapse images of the dynamic events in yellow-boxed regions (c). e, The y-t plots corresponding to the trajectories indicated in d. Orange arrows indicate the 256 s and 300 s time points in the sequence. All experiments were repeated independently for more than 3 ROIs with similar results. Gamma, 0.7 (a-e). Scale bars, 2 μm (a,b), 5 μm (c), and 1 μm (d).
Extended Data Fig. 9 SSR-SIM imaging of actin comet tails in Vero cells infected with Vaccinia Virus (VACV).
a, Vero cells stably expressing LifeAct-mRuby (Orange Hot) and stained with Hoechst dyes (Cyan Hot). Experiments were repeated independently for 3 cells with similar results. b-d, Time-lapse images of VACV particles and actin comet tails. e, Movement trajectory of the VACV particle indicated in d. The white arrow denotes the position at the 94-second time point. Gamma, 0.7 (a-d). Scale bars, 5 μm (a), 2 μm (b-d).
Supplementary information
Supplementary Information (download PDF )
Supplementary Notes 1–9, Figs. 1–21 and Tables 1–4.
Supplementary Video 1 (download MP4 )
Comparison of SSR-TIRF-SIM and other SIM algorithms on long-term imaging of actin cytoskeleton.
Supplementary Video 2 (download MP4 )
Comparison of SSR-TIRF-SIM and other SIM algorithms on imaging of F-actin’s remodeling dynamics during the adhesion process after dropping cells onto the coverslip.
Supplementary Video 3 (download MP4 )
SSR-TIRF-SIM facilitates high-speed live-cell imaging of F-actin exceeding 10,000 frames.
Supplementary Video 4 (download MP4 )
Comparison of SSR-GI-SIM and other SIM algorithms on time-lapse imaging of microtubules in COS-7 cells.
Supplementary Video 5 (download MP4 )
Comparison of SSR-GI-SIM and other SIM algorithms on time-lapse imaging of mitochondrial dynamics in COS-7 cells.
Supplementary Video 6 (download MP4 )
SSR-GI-SIM reveals ER and VSVG interactions with high spatiotemporal resolution.
Supplementary Video 7 (download MP4 )
SSR-GI-SIM reveals rapid movement of VSVG leads to the remodeling of ER sheets.
Supplementary Video 8 (download MP4 )
SSR-GI-SIM reveals deformable morphologies of the Rab11a and TfR1 complex during endocytic recycling.
Supplementary Video 9 (download MP4 )
SSR-TIRF-SIM demonstrates VACV-induced dual and multiple actin comet tails to aid its movement and propagation.
Supplementary Video 10 (download MP4 )
SSR-GI-SIM reveals mitochondrial fission events occurring in intercellular tunneling nanotubes.
Supplementary Video 11 (download MP4 )
SSR-3D-SIM imaging of the inner mitochondrial membrane with over 300 frames acquired.
Source data
Source Data Fig. 1 (download XLSX )
Statistical source data.
Source Data Fig. 2 (download XLSX )
Line profile source data.
Source Data Fig. 5 (download XLSX )
Statistical source data.
Source Data Extended Data Fig. 1 (download XLSX )
Statistical source data.
Source Data Extended Data Fig. 3 (download XLSX )
Statistical source data.
Source Data Extended Data Fig. 7 (download XLSX )
Statistical source data.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Liu, J., Dong, X., Lu, H. et al. Bio-friendly and high-precision super-resolution imaging through self-supervised reconstruction structured illumination microscopy. Nat Methods 23, 395–404 (2026). https://doi.org/10.1038/s41592-025-02966-y
Received:
Accepted:
Published:
Version of record:
Issue date:
DOI: https://doi.org/10.1038/s41592-025-02966-y
