Automatic segmentation of mitochondria and endolysosomes in volumetric electron microscopy data

Area of Science:

• Cell Biology
• Microscopy
• Computational Biology

Background:

• Automatic segmentation of intracellular compartments provides crucial quantitative data on cell presence, distribution, structure, and function.
• High-throughput volumetric data acquisition in electron microscopy (EM) has made manual segmentation a significant bottleneck.

Purpose of the Study:

• To develop an automated segmentation technique for mitochondria and endolysosomes in urothelial cells using dual-beam EM.
• To introduce a novel, publicly available volumetric EM dataset of urothelial cells.
• To evaluate existing segmentation methods and present an improved pipeline.

Main Methods:

• A novel, publicly available volumetric EM dataset of urothelial cells was created.
• Several state-of-the-art segmentation methods were evaluated on the new dataset.
• A supervised deep learning segmentation pipeline was developed, incorporating mechanisms to mitigate data dependencies, artifacts, and annotation errors.

Main Results:

• The proposed deep learning segmentation pipeline demonstrated superior performance compared to existing methods on the urothelial cell EM dataset.
• The novel dataset provides a valuable resource for advancing research in urothelial cell ultrastructure.

Conclusions:

• The developed automated segmentation technique effectively addresses the limitations of manual segmentation in EM data.
• This approach enhances the efficiency and accuracy of quantitative analysis of intracellular compartments in urothelial cells.