Fig. 1. Ion-beam image reconstruction framework achieves high-precision imaging with multiplexing.

a Metabolic labeling or metal-tagging in ion beam imaging. Continuous ion beam scans of up to 1000 depth sections were performed across the entire cell. b Reconstruction methods of serial image arrays from ion beam tomography (IBT) scans. Ion images-Sum: Conventional approach in SIMS image analysis, a range of 5-20 slices were directly summed. Deconvolution-IBT: A mathematical deblurring routine that iteratively removes out-of-focus etching. SILM-IBT: Localization of isolated ion molecules at each depth scan and synthesis of a highly precise ion image from localized ion distributions across 20–100 slices. Deep-learning-IBT: Training a neural net model on raw image stacks (low resolution) to SILM reconstructed image arrays (high precision) for deep learning-based reconstruction method to create a predicted image stack. c Co-tagged ¹²⁷I-dU and ⁸¹Br-dU in a HeLa cell for chromatin mapping. (Top row) ¹²⁷I channel in the SIMS data for newly synthesized DNA (¹²⁷I-dU) after 24 h incorporation. Deconvolution resolves some of the overlapping chromatin features, while SILM and Deep-learning reconstructions finely resolve sub-25 nm chromatin fibers. The inset shows a cross-section across the DNA. Scale bars, 2 µm. (Bottom row) ⁸¹Br channels for the same DNA but now labeled with ⁸¹Br-dU simultaneously incubated with ¹²⁷I-dU. Ion image arrays showed resolution enhancement from deconvolution to SILM and Deep-learning analysis, as before. d Comparison of image reconstruction methods by analyzing isotope encoded nanotags (100 nm diameters). ¹⁹F isotope was uniformly embedded in a Silica (²⁸Si) matrix. (Top row) ²⁸Si channel in the SIMS images for nanotags. The sum of the frames and Deconvolution-IBT resolve decent ion beam overlaps, and SILM and deep learning analysis allow sub-ion-beam-width reconstructions. Inset shows a cross-section of two neighboring nanotags, denoted by an arrow. Scale bars, 1 µm. (Bottom row) ¹⁹F channel in the same nanotags as a colocalized multi-color visuals. The SILM-IBT ¹⁹F and ²⁸Si patterns agree well with each other with Structural similarity index (SSIM) values of 0.7. Reconstructions from Deep-learning-IBT and SILM provided SSIM values of 0.66. Inset corresponds to the resolved nanotag pair. Credit: Designua, Timonina, and Alejo Miranda/Shutterstock.com.