An automated pipeline for mitochondrial segmentation on ATUM-SEM stacks

Area of Science:

• Cell Biology
• Neuroscience
• Microscopy Imaging

Background:

• Electron microscopy (EM) enables detailed visualization of mitochondrial structures crucial for cellular and neuronal functions.
• Accurate segmentation of mitochondria in EM images is challenging due to complex subcellular environments and image artifacts.
• Current algorithms struggle with segmenting mitochondria near vesicles or membranes.

Purpose of the Study:

• To develop an improved algorithm for segmenting mitochondria in electron microscopy images.
• To address limitations of existing methods in segmenting closely packed or adjacent mitochondria.

Main Methods:

• Explicitly modeling mitochondrial double membrane structures.
• Utilizing ridge detection for acquiring image edges instead of image gradients.
• Applying group-similarity for optimizing local segmentation accuracy.

Main Results:

• The proposed algorithm demonstrates enhanced performance in segmenting mitochondria from EM images.
• Effectiveness validated on images acquired using automated tape-collecting ultramicrotome scanning electron microscopy (ATUM-SEM).
• Improved segmentation accuracy, particularly for mitochondria in challenging proximity to other structures.

Conclusions:

• The novel approach effectively segments mitochondria by focusing on membrane structures and advanced edge detection.
• This method offers a more robust solution for mitochondrial segmentation in complex biological imaging.
• The findings contribute to advancing quantitative analysis of mitochondrial morphology in cellular and neuronal studies.