Deep learning empowers photothermal microscopy with super-resolution capabilities.

Area of Science:

• Optics and Photonics
• Microscopy Techniques
• Artificial Intelligence in Imaging

Background:

• Photothermal microscopy (PTM) offers high sensitivity for single-particle/molecule detection in materials science and biology.
• PTM's far-field nature imposes diffraction-limited resolution.
• Previous methods like modulated difference PTM (MDPTM) showed limited resolution improvement due to artifacts and information loss.

Purpose of the Study:

• To develop a deep learning approach for significantly enhancing the lateral resolution of PTM.
• To overcome the resolution constraints of conventional and modulated difference PTM.
• To introduce a novel super-resolution microscopy technique for nanoscale imaging.

Main Methods:

• A deep learning approach using Cycle Generative Adversarial Network (Cycle GAN) was employed, termed DMDPTM.
• Optimized point spread functions (PSFs) of PTM and MDPTM were integrated as a second generator in the Cycle GAN.
• Dataset construction utilized the relationship between sample volume and photothermal signal, incorporating both PTM and MDPTM images as inputs.

Main Results:

• Simulations demonstrated DMDPTM resolving 60 nm distances between 60 nm nanoparticles, a 4.4-fold resolution enhancement over conventional PTM.
• Experimental validation on gold nanoparticles achieved a resolution of 114 nm.
• Successful imaging of carbon nanotubes was demonstrated, showcasing practical applicability.

Conclusions:

• DMDPTM effectively improves the lateral resolution of PTM beyond diffraction limits.
• This deep learning-based method offers a powerful tool for high-resolution imaging of nanoscale structures.
• The technique holds significant potential for applications in materials science and biology requiring nanoscale resolution.