Whole-cell organelle segmentation in volume electron microscopy.

Area of Science:

• Cell Biology
• Biophysics
• Computational Biology

Background:

• Understanding cellular organization requires high-resolution 3D reconstruction of organelles.
• Scalable automated methods are crucial for achieving whole-cell reconstructions.

Purpose of the Study:

• To develop and validate automated deep learning methods for segmenting and reconstructing cellular organelles from FIB-SEM data.
• To create an open-access resource for 3D cell biology research.

Main Methods:

• Annotation of 35 organelle classes in FIB-SEM datasets (4nm and 8nm resolution).
• Training deep learning models for segmentation of cellular structures.
• Validation of segmentation performance and application in quantitative analysis.

Main Results:

• Successful segmentation and 3D reconstruction of diverse organelles from multiple cell types.
• Demonstration of automated quantification of organelle spatial interactions.
• Development of automatic registration for correlative light and electron microscopy.

Conclusions:

• Automated 3D reconstruction methods significantly advance the study of cellular architecture.
• The OpenOrganelle repository facilitates global access to data, code, and models for improved cell reconstruction.
• This work enables new quantitative analyses of cellular organization and correlative imaging.