Fourier-Based 3D Multistage Transformer for Aberration Correction in Multicellular Specimens

Area of Science:

• Microscopy
• Biophysics
• Machine Learning

Background:

• Optical aberrations degrade image quality in high-resolution tissue imaging.
• Conventional adaptive optics (AO) solutions are complex, costly, and slow for large-scale aberration correction.

Purpose of the Study:

• To introduce AOViFT (Adaptive Optical Vision Fourier Transformer), a novel machine learning framework for aberration sensing and correction.
• To reduce the computational cost and complexity associated with aberration correction in microscopy.

Main Methods:

• Developed a 3D multistage Vision Transformer operating on Fourier domain embeddings.
• Implemented AOViFT for aberration inference and restoration of diffraction-limited performance.
• Validated the framework on live gene-edited zebrafish embryos.

Main Results:

• AOViFT achieved aberration correction with reduced computational cost, training time, and memory footprint compared to traditional methods.
• Demonstrated successful correction of spatially varying aberrations using both hardware (deformable mirror) and software (post-acquisition deconvolution) approaches.
• Validated performance on live biological samples, showcasing its practical applicability.

Conclusions:

• AOViFT offers an efficient, AI-driven solution for correcting optical aberrations in microscopy.
• The framework simplifies experimental workflows by eliminating the need for guide stars and wavefront sensing hardware.
• AOViFT lowers technical barriers, enhancing accessibility to high-resolution volumetric microscopy for diverse biological applications.