Related Experiment Video
Updated: Nov 7, 2025
10:25
Deep Learning-Based Segmentation of Cryo-Electron Tomograms
Published on: November 11, 2022
9.9K
Deep learning for automatic segmentation of the nuclear envelope in electron microscopy data, trained with volunteer
Helen Spiers1,2, Harry Songhurst1,3, Luke Nightingale4
1Electron Microscopy Science Technology Platform, The Francis Crick Institute, London, UK.
Traffic (Copenhagen, Denmark)
|April 29, 2021
Summary
Citizen scientists in the Etch a Cell project manually segmented cell structures. This data successfully trained a machine learning algorithm for automated nuclear envelope segmentation, overcoming a key research bottleneck.
Area of Science:
- Cell Biology
- Microscopy
- Machine Learning
Background:
- Volume electron microscopy generates large datasets, but manual segmentation by experts is a bottleneck.
- Existing machine learning methods lack sufficient high-quality training data for accurate, generic automated analysis.
Purpose of the Study:
- To develop a citizen science approach for generating high-quality ground-truth data for electron microscopy.
- To train a machine learning algorithm for automated nuclear envelope segmentation using volunteer-generated data.
Main Methods:
- A novel citizen science project, Etch a Cell, was created for volunteers to segment the nuclear envelope (NE) in serial blockface scanning electron microscopy data.
- Multiple volunteer annotations were aggregated to create a consensus segmentation.
- The consensus data was used to train a machine learning algorithm.
Main Results:
- Volunteer-annotated data was sufficient to train a highly accurate machine learning algorithm for NE segmentation.
- The developed algorithm enables automated analysis of electron microscopy data.
Conclusions:
- Citizen science can effectively generate high-quality training data for machine learning in microscopy.
- Automated NE segmentation is achievable, significantly advancing research in cell biology and microscopy data analysis.
