Patterned illumination enables denser deep-learning based single-molecule localization microscopy

Abstract

Single-molecule localization microscopy (SMLM) improves resolution beyond diffraction limits by imaging cellular structures at the nanometer scale. Recent advances include using modulation patterns to improve localization precision, and deep learning to accurately process high-density samples with overlapping fluorophore emissions, thus improving imaging speed. A method combining these two advances, SIMCODE, is presented here, allowing high-density modulation-enhanced SMLM. SIMCODE achieved resolution improvements in high-density areas compared to SMLM, deep learning-based SMLM (DECODE), and modulation-enhanced SMLM alone (SIMFLUX). In DNA-PAINT imaging of COS-7 cells, SIMCODE showed improvements in the Fourier Ring Correlation and resolution-scaled Pearson coefficient, with overall improvement increasing as imaging buffer concentration increased five-fold. Modulation-enhanced localization microscopy combined with deep learning thus produced higher quality reconstructions at higher emitter densities (i.e., [~]3x the number of detected spots). This will enable faster imaging, higher labeling densities, and more flexibility in fluorophore choice, which are important for studying dynamic processes and densely labeled structures.