Mapping mesoscale axonal projections in the mouse brain using a 3D convolutional network.

Area of Science:

• Neuroscience
• Computational Biology
• Bioimaging

Background:

• Traditional methods for mapping neuronal projections are labor-intensive and provide coarse resolution.
• Understanding mesoscale projectomes is crucial for neural circuit analysis.

Purpose of the Study:

• To introduce TrailMap, a 3D convolutional network for automated extraction and quantification of axonal projections.
• To enable detailed analysis of neuronal connectivity in cleared mouse brains.

Main Methods:

• Utilized a 3D convolutional neural network (TrailMap) for axonal projection extraction from light-sheet microscopy data.
• Employed data augmentation and a specialized loss function for accurate identification of fine axonal structures.
• Applied transfer learning to adapt the model for novel axonal morphologies.

Main Results:

• TrailMap accurately quantifies total axon content in large, complex 3D brain structures after registration to a reference atlas.
• The network demonstrates generalization capabilities across different neuronal projection types (serotonergic, thalamocortical, cerebellar, cortical).
• TrailMap offers improved speed, ease of use, and accuracy compared to existing tools without requiring specialized hardware.

Conclusions:

• TrailMap facilitates automated, whole-brain scale extraction and quantification of axons from specific cell types.
• This advancement is essential for deciphering neural connectivity and understanding neural circuits.
• The tool supports the growing emphasis on genetically and functionally defined cell types in neuroscience research.