Related Experiment Video
Updated: Nov 22, 2025
12:08
From Voxels to Knowledge: A Practical Guide to the Segmentation of Complex Electron Microscopy 3D-Data
Published on: August 13, 2014
24.8K
Sequential semi-supervised segmentation for serial electron microscopy image with small number of labels.
Eichi Takaya1, Yusuke Takeichi2, Mamiko Ozaki3
1School of Science for Open and Environmental Systems, Graduate School of Science and Technology, Keio University, Kanagawa, Japan.
Journal of Neuroscience Methods
|January 8, 2021
Summary
This study introduces sequential semi-supervised segmentation (4S) for electron microscopy images. The 4S method effectively segments neural regions using limited labeled data and unlabeled data, outperforming traditional supervised learning.
Area of Science:
- Neuroscience
- Computational Biology
- Image Analysis
Background:
- Deep learning for electron microscopy image segmentation aids 3D reconstruction.
- Generalization to new samples is challenging with rare samples or unstable scanning.
Purpose of the Study:
- To develop a method for semi-automatic neural region extraction from electron microscopy image stacks.
- To address limitations of generalization performance in transductive settings.
Main Methods:
- Proposed sequential semi-supervised segmentation (4S) in a transductive setting.
- Leveraged correlations between adjacent serial images.
- Iterative training, inference, and pseudo-labeling with minimal teacher labels.
Main Results:
- Demonstrated effectiveness in both quality and quantity on two types of serial section images.
- Clarified the impact of teacher label quantity and position on segmentation performance.
- Achieved superior performance compared to supervised learning with limited labeled data.
Conclusions:
- 4S effectively extracts neural regions from serial image stacks using limited labeled and abundant unlabeled data in a transductive setting.
- The method shows promise as a core component for future annotation tools.
