Correcting Non-Uniform Milling in FIB-SEM Images with Unsupervised Cross-Plane Image-to-Image Translation

Area of Science:

• * Advanced microscopy techniques
• * Biological imaging and analysis
• * Computational imaging

Background:

• * Focused Ion Beam Scanning Electron Microscopy (FIB-SEM) offers high axial resolution for 3D biological structure analysis.
• * Non-uniform milling thickness in FIB-SEM causes axial plane distortions, hindering accurate segmentation and reconstruction.
• * Current image processing methods often assume uniform thickness, lack end-to-end processing, or require laborious ground truth data.

Purpose of the Study:

• * To develop an advanced deep learning method for correcting non-uniform milling artifacts in FIB-SEM images.
• * To mitigate image distortions in an unsupervised manner, eliminating the need for ground truth annotations.
• * To improve the accuracy and efficiency of 3D biological structure analysis using FIB-SEM data.

Main Methods:

• * Implemented an unsupervised image-to-image translation deep learning technique.
• * Utilized cross-plane learning within 3D image volumes to address distortions.
• * Applied the method to a real-world micro-wasp dataset without requiring ground truth images.

Main Results:

• * Successfully corrected non-uniform milling artifacts in FIB-SEM images.
• * Demonstrated significant improvements in image quality through qualitative and quantitative analysis.
• * Validated the efficacy of the unsupervised deep learning approach on complex biological samples.

Conclusions:

• * The developed deep learning method effectively corrects non-uniform milling artifacts in FIB-SEM imaging.
• * Unsupervised, cross-plane learning offers a robust solution for distortion mitigation without ground truth data.
• * This approach enhances the reliability of 3D reconstruction and analysis of fine biological structures.