Efficient semi-automatic 3D segmentation for neuron tracing in electron microscopy images.

Area of Science:

• Connectomics
• Neuroscience
• Electron Microscopy Image Analysis

Background:

• Significant time gap exists between data acquisition and neural process reconstruction in connectomics.
• Current automatic methods lack sufficient accuracy for direct use, requiring tedious manual corrections.
• Manual correction of neural reconstructions is time-consuming and labor-intensive.

Purpose of the Study:

• To develop a semi-automatic method for efficient and accurate 3D segmentation of neurites in electron microscopy image stacks.
• To reduce the time and effort required for correcting automatically segmented neural data.
• To improve the usability of automated connectomics reconstruction tools.

Main Methods:

• A novel semi-automatic method utilizing a hierarchical structure for superpixel merges is presented.
• The method guides users through predicted regions, facilitating rapid error identification and correction.
• User interaction is focused on verifying and establishing correct links within the segmented neural data.

Main Results:

• The method was evaluated on three electron microscopy datasets with both novice and expert users.
• Accuracy and timing were benchmarked against existing automatic, semi-automatic, and manual segmentation techniques.
• Novice users achieved accuracy surpassing state-of-the-art automatic methods.

Conclusions:

• The developed semi-automatic method significantly improves the efficiency of neural process segmentation.
• Expert users achieved manual-level accuracy with a 70% reduction in time compared to previous methods.
• This approach bridges the gap between automated reconstruction speed and manual accuracy in connectomics.