Deep learning-based aberration compensation improves contrast and resolution in fluorescence microscopy.

Area of Science:

• Biophysics
• Microscopy
• Computational Biology

Background:

• Optical aberrations degrade fluorescence microscopy images of thick samples, limiting signal, contrast, and resolution.
• Existing aberration correction methods may require additional optics, increase radiation dose, or slow image acquisition.

Purpose of the Study:

• To develop a deep learning-based strategy for efficient and effective optical aberration compensation in fluorescence microscopy.
• To improve image quality and downstream quantitative analysis without compromising acquisition speed or sample integrity.

Main Methods:

• A deep learning approach was developed involving the introduction of synthetic aberrations to shallow image planes.
• Neural networks were trained to reverse the effects of these synthetic aberrations, creating 'de-aberration' networks.
• The method was validated using simulations and experiments across various microscopy techniques.

Main Results:

• The deep learning 'de-aberration' networks significantly outperformed alternative methods in correcting optical aberrations.
• Restored image quality was comparable to that achieved with adaptive optics techniques.
• The method successfully improved qualitative inspection and quantitative analysis in diverse microscopy datasets.

Conclusions:

• Deep learning-based aberration compensation offers a powerful, non-invasive solution for enhancing fluorescence microscopy of thick samples.
• This approach facilitates improved biological insights through better image quality and more accurate quantitative measurements.
• The method is broadly applicable across various fluorescence microscopy modalities, including confocal, light-sheet, multi-photon, and super-resolution imaging.