Scalable and interactive segmentation and visualization of neural processes in EM datasets

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Imaging

Background:

• High-resolution electron microscopy (EM) generates vast datasets for neural circuit reconstruction.
• Manual segmentation and visualization of neural processes in EM data are time-consuming and complex.

Purpose of the Study:

• To present NeuroTrace, a novel system for EM volume segmentation and visualization.
• To improve the efficiency and ease of reconstructing complex neural processes from EM data.

Main Methods:

• Developed a semi-automatic multiphase level set segmentation with 3D tracking for neural process reconstruction.
• Implemented a specialized volume rendering approach with view-dependent filtering and on-the-fly interpolation.
• Utilized GPU acceleration for interactive performance and scalability to large datasets.

Main Results:

• NeuroTrace offers faster and easier reconstruction of complex neural processes compared to manual methods.
• The system provides interactive performance for segmentation and visualization tasks.
• Demonstrated scalability for large datasets and parallel hardware architectures.

Conclusions:

• NeuroTrace significantly enhances the workflow for neural process reconstruction in EM data.
• The system's interactive and efficient approach aids neuroscientists in analyzing complex neural connectomes.
• GPU-accelerated techniques are crucial for handling large-scale EM datasets.