An automated images-to-graphs framework for high resolution connectomics

Area of Science:

• Neuroscience
• Computational Neuroscience
• Bioinformatics

Background:

• Mapping neuronal connectivity is crucial for understanding brain function.
• High-throughput electron microscopy (EM) generates large 3D brain tissue datasets.
• Current automated methods address specific tasks, not end-to-end reconstruction.

Purpose of the Study:

• To present the first fully-automated images-to-graphs pipeline for reconstructing neuronal networks.
• To develop a metric for evaluating the quality of reconstructed brain graphs.
• To establish a baseline for end-to-end pipeline performance on public data.

Main Methods:

• Developed a fully-automated pipeline from 3D EM image volumes to brain graphs.
• Created a novel metric to assess the quality of output brain graphs.
• Evaluated various algorithms and parameters to optimize pipeline performance.

Main Results:

• Successfully generated brain graphs from EM data without human intervention.
• Identified optimal algorithms and parameters using the developed quality metric.
• Deployed a reference end-to-end pipeline on a large public dataset.

Conclusions:

• The presented pipeline offers a scalable solution for reconstructing neuronal connectivity.
• The developed metric and baseline results facilitate community-driven advancements.
• Publicly released code and data support future research in connectomics and neuropathology.