Universal adaptive optics for microscopy through embedded neural network control.

Area of Science:

• Optical microscopy
• Biomedical imaging
• Machine learning applications

Background:

• Microscope imaging resolution and contrast are limited by optical aberrations.
• Existing adaptive optics (AO) solutions are often modality-specific.
• A universal AO system for diverse microscopes is needed.

Purpose of the Study:

• To develop a versatile and fast aberration correction method for microscopy.
• To create a physics-based, machine learning assisted wavefront-sensorless AO control (MLAO) system.
• To enable AO transferability across different microscope modalities.

Main Methods:

• Implemented a physics-based machine learning assisted wavefront-sensorless AO control (MLAO) method.
• Designed a novel neural network (NN) architecture based on physical understanding of image formation.
• Embedded the NN in the control loop of two-photon, three-photon, and widefield 3D structured illumination microscopes.

Main Results:

• The MLAO method demonstrated faster and more effective aberration correction than traditional modal-based sensorless AO.
• Successfully applied the method across diverse microscopy techniques, including two-photon, three-photon, and widefield 3D structured illumination.
• Showcased robustness in challenging conditions like 3D structures, specimen motion, low signal-to-noise ratio, and fluorescence fluctuations.

Conclusions:

• The developed MLAO method offers a universal, translatable solution for aberration correction in microscopy.
• The physics-informed NN provides physical insights, moving beyond a 'black box' approach.
• This approach enhances image quality and robustness in complex biological imaging scenarios.