Three-Dimensional Reconstruction of Serial Block-Face Scanning Electron Microscopy Using Semantic Segmentation based...

Area of Science:

• Electron Microscopy
• Computational Biology
• Machine Learning

Background:

• Serial block-face scanning electron microscopy (SBF-SEM) enables high-resolution 3D reconstructions of cellular ultrastructures.
• Accurate segmentation of SBF-SEM data is crucial for reliable analysis but is labor-intensive.

Purpose of the Study:

• To develop a semi-supervised learning approach for efficient and accurate segmentation in SBF-SEM.
• To reduce the manual annotation effort required for SBF-SEM data processing.

Main Methods:

• A novel segment interpolation method was used to estimate segment shapes and locations between sparsely annotated images.
• Deep neural networks were trained using this semi-supervised approach.
• Validation employed F-1 scores and K-fold cross-validation.

Main Results:

• The method achieved an F-1 score of 0.89 for mouse brain cells and 0.84 for inverted images during validation.
• Testing on an independent dataset yielded scores of 0.84 (mouse brain cells) and 0.80 (inverted cases).
• Reconstruction using the marching cube algorithm enabled 3D analysis of organelles.

Conclusions:

• The proposed semi-supervised learning method significantly improves SBF-SEM segmentation accuracy while minimizing manual annotation.
• This approach facilitates detailed 3D ultrastructural analysis of complex organelles.
• Potential applications exist in biology and medicine for advanced cellular imaging and analysis.