Surpassing the optical diffraction limit through label-free illumination diversity and neural network reconstruction

Session

Advances in Label-free Imaging

Authors

James Lillis (1), Ning Liu (1), Syed A.M. Tofail (1), Christophe Silien (1)

Affiliations

1. University of Limerick

Keywords

Super-resolution

Label-free Imaging

Machine Learning

Neural Networks

Vibrational Imaging

Illumination Diversity

Abstract text

Light microscopes have a rich history in the life sciences and are uniquely suited to image and characterise delicate biological samples in their natural environments. Optical instruments, however, are limited in resolution by the diffraction of light to several hundred nanometres, preventing the application of valuable optical techniques to nanoscale features and samples. Developments in optical super-resolution techniques over the past three decades have broken past the classical diffraction limit, with modern fluorescence microscopes readily achieving resolutions of several nanometres. Current fluorescence approaches rely on the use of delicate fluorophore dyes with numerous well-known drawbacks including photobleaching, limited lifetimes and photostability, fluorophore mediated effects, and difficult multiplexing. Label-free microscopy modalities don’t depend on the use of fluorescent dyes avoid these drawbacks while providing unique insights into the chemical structure of biological samples. Vibrational techniques like Raman and Infrared microscopy provide valuable spectro-spatial information in the fingerprint region, and nonlinear techniques like coherent raman (CARS & SRS) and sum frequency generation (SHG, THG, etc) provide label-free chemical contrast in a diverse range of samples. While label-free super-resolution microscopes exist, they are limited generally to a 2x improvement over the diffraction limit, falling well short of what labelled super-resolution can achieve. 

Here, we demonstrate a label-free optical super-resolution technique based on conventional image scanning microscopy that incorporates illumination diversity and neural network image reconstruction to surpass the diffraction limit. Our approach is suitable for low power illumination (repeated imaging of delicate biological samples), is broadly applicable to label-free modalities, can be easily incorporated to existing microscope platforms, and can out-perform conventional SIM and ISM approaches in resolution. We demonstrate the applicability of our approach through computational simulations and experimental validation.

References

1. Silva, W. R., Graefe, C. T. & Frontiera, R. R. Toward Label-Free Super-Resolution Microscopy. ACS Photonics 3, 79–86 (2016). 

2. Graefe, C. T. et al. Far-Field Super-Resolution Vibrational Spectroscopy. Analytical Chemistry 91, 8723–8731 (2019). 

3. Leighton, R. E., Alperstein, A. M. & Frontiera, R. R. Label-Free Super-Resolution Imaging Techniques. Annual Review of Analytical Chemistry 14, 39 (2022)