Deep learning-based autofocus method enhances image quality in light-sheet fluorescence microscopy.

Area of Science:

• Biomedical Imaging
• Microscopy Techniques
• Computational Biology

Background:

• Light-sheet fluorescence microscopy (LSFM) enables high-throughput, high-resolution imaging of large biological samples.
• Accurate alignment between the light-sheet and objective focal planes is critical for LSFM.
• Refractive index variations and spherical aberrations deep within tissues disrupt this alignment, hindering imaging quality.

Purpose of the Study:

• To develop a rapid and robust autofocus method for LSFM to overcome alignment challenges.
• To improve the accuracy and efficiency of focusing in deep-tissue LSFM.
• To enable high-resolution imaging of large 3D specimens.

Main Methods:

• A deep learning framework was developed to estimate objective lens focal plane position using two defocused images.
• The deep learning model was trained and validated on image patches.
• The autofocus framework was integrated into a custom-built LSFM system with a certainty measure for refinement.

Main Results:

• The deep learning autofocus method demonstrated performance comparable to or exceeding traditional methods on various image patch sizes.
• Real-time autofocusing was achieved on cleared mouse forebrain and pig cochleae samples.
• The system successfully maintained focus during imaging of complex biological specimens.

Conclusions:

• Deep learning-based autofocus provides a significant advancement for light-sheet fluorescence microscopy.
• This framework enhances the capability of LSFM for imaging large, 3D biological structures with sub-cellular detail.
• The developed method has broad implications for advancing biological research through improved microscopy.