Fourier-based three-dimensional multistage transformer for aberration correction in multicellular specimens.

Area of Science:

• Microscopy and Imaging Science
• Artificial Intelligence in Biology
• Optical Physics

Background:

• High-resolution tissue imaging is limited by optical aberrations, reducing image quality.
• Conventional adaptive optics (AO) solutions are complex, costly, and slow for large-scale imaging.

Purpose of the Study:

• To introduce AOVIFT, a novel machine learning framework for aberration sensing and correction in microscopy.
• To reduce the computational cost and hardware requirements for high-resolution imaging.

Main Methods:

• Developed AOVIFT, a machine learning framework using a 3D multistage vision transformer operating on Fourier domain embeddings.
• Inferred optical aberrations and restored diffraction-limited performance in puncta-labeled specimens.
• Validated on live gene-edited zebrafish embryos, correcting aberrations with a deformable mirror or post-acquisition deconvolution.

Main Results:

• AOVIFT achieved aberration correction with reduced computational cost, training time, and memory footprint compared to existing methods.
• Successfully demonstrated correction of spatially varying aberrations in live biological samples.
• Enabled high-resolution volumetric microscopy with a simplified experimental workflow.

Conclusions:

• AOVIFT effectively corrects optical aberrations, enhancing resolution and contrast in microscopy.
• The AI-driven approach eliminates the need for wavefront sensing hardware, lowering technical barriers.
• AOVIFT facilitates broader access to high-resolution volumetric imaging across various biological applications.