Axon Tracing and Centerline Detection using Topologically-Aware 3D U-Nets.

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Accurate reconstruction of neural connectivity is crucial for understanding brain structure-function relationships.
• Manual tracing of neural networks is time-consuming, labor-intensive, and requires specialized expertise.
• Automated methods are essential for efficient and scalable analysis of neural connectivity.

Purpose of the Study:

• To explore deep neural networks for automated dense axon tracing.
• To improve voxel-based segmentation and axon centerline detection by incorporating topological information.
• To enhance the accuracy and efficiency of neural connectivity reconstruction.

Main Methods:

• Utilized deep neural networks, specifically a modified 3D U-Net architecture.
• Trained models on a mouse brain dataset acquired via light sheet microscopy.
• Incorporated axon topological information into the loss function to guide the network.

Main Results:

• Achieved a 10% increase in axon tracing accuracy compared to previous methods.
• The addition of centerline awareness in the loss function significantly improved performance across all metrics.
• Observed an 8% boost in Rand Index, indicating enhanced segmentation accuracy.

Conclusions:

• Deep neural networks, enhanced with topological information, offer a powerful solution for automated dense axon tracing.
• The developed method significantly improves the accuracy and efficiency of neural connectivity reconstruction.
• This approach facilitates large-scale analysis of brain structure and function.