Machine learning-based 3D segmentation of mitochondria in polarized epithelial cells

Area of Science:

• Cell Biology
• Mitochondrial Dynamics
• Bioimaging

Background:

• Mitochondrial morphology reflects cellular health and function.
• Accurate segmentation of mitochondrial networks is essential for quantitative analysis.
• 3D segmentation of complex mitochondrial networks, especially in polarized cells, is challenging.

Purpose of the Study:

• To develop and validate a machine learning approach for improved 3D mitochondrial segmentation.
• To enhance the accuracy of mitochondrial network segmentation in super-resolution microscopy images.
• To apply quantitative analysis to study mitochondrial morphology changes in specific cellular models.

Main Methods:

• Utilized a machine learning approach with the 3D Trainable Weka plugin for ImageJ.
• Applied the method to super-resolution microscopy images of mitochondrial networks.
• Tested the approach on differentiated human retinal pigment epithelial (RPE) cells and other polarized epithelial cell models.

Main Results:

• The machine learning approach significantly improved the accuracy of 3D mitochondrial segmentation compared to existing methods.
• Effective segmentation of complex mitochondrial networks was achieved in various polarized cell types.
• Quantitative analysis revealed mitochondrial fragmentation in RPE cells treated with bafilomycin.

Conclusions:

• Machine learning, specifically 3D Trainable Weka, offers a powerful solution for accurate 3D mitochondrial segmentation.
• This method enhances the study of mitochondrial dynamics and cellular health in complex biological systems.
• The findings provide a reliable tool for researchers investigating mitochondrial function and disease.