A generalist deep-learning volume segmentation tool for volume electron microscopy of biological samples

Area of Science:

• * Computational Biology
• * Neuroscience
• * Materials Science

Background:

• * Volumetric image segmentation is crucial for analyzing complex biological structures in 3D.
• * Existing methods often require significant manual intervention and computational resources.

Purpose of the Study:

• * To develop an accessible, automated deep learning tool for volumetric image segmentation.
• * To adapt the tool for diverse biological sample types and electron microscopy datasets.

Main Methods:

• * Developed the Volume Segmentation Tool (VST) using deep learning.
• * Implemented automated data preprocessing, augmentation, network building, and model training.
• * Integrated a browser-based interface for local hardware operation and visualization.

Main Results:

• * VST successfully performs semantic and instance segmentation using contour map prediction.
• * Demonstrated state-of-the-art performance on transmission electron microscopy and scanning electron microscopy datasets.
• * Achieved high accuracy across various resin-embedded biological samples.

Conclusions:

• * VST provides an automated and accessible solution for volumetric image segmentation.
• * The tool enhances the analysis of large-scale 3D electron microscopy data.
• * VST represents a significant advancement in biological image analysis software.