Computerized detection and segmentation of mitochondria on electron microscope images.

Area of Science:

• Cell Biology
• Biophysics
• Computational Biology

Background:

• Mitochondrial dysfunction is implicated in various diseases.
• Electron microscopy tomography (EMT) reveals 3D mitochondrial structure.
• Automated segmentation of mitochondria in EMT data is challenging due to artifacts and complex subcellular structures.

Purpose of the Study:

• To develop computer algorithms for automatic detection and segmentation of mitochondria in electron microscopy images.
• To address the time-consuming nature of manual segmentation for high-throughput analysis.
• To improve the accuracy and efficiency of analyzing mitochondrial distribution and features.

Main Methods:

• Utilized mitochondria's elliptical shape and double membrane for initial detection.
• Employed active contours for refining detection results.
• Implemented a seed point selection method combined with a live-wire graph search algorithm for final segmentation.

Main Results:

• Achieved 91% Dice similarity coefficient and an average 4.9 nm boundary error compared to manual segmentation.
• Successfully detected 14 out of 15 fully visible mitochondria and 4 out of 7 partially visible mitochondria.
• Demonstrated accurate detection and segmentation of both complete and partial mitochondria.

Conclusions:

• The developed algorithms provide an effective solution for automated mitochondrial detection and segmentation in electron microscopy.
• This advancement facilitates high-throughput analysis of mitochondrial morphology and distribution.
• The method shows promise for accelerating research in cellular events and disease states involving mitochondria.