Connectome quality converges predictably to reveal optimal stopping points during proofreading

Area of Science:

• Neuroscience
• Computational Biology
• Microscopy

Background:

• Volumetric electron microscopy (EM) generates high-resolution brain tissue reconstructions.
• Automated segmentation in EM pipelines introduces errors requiring manual proofreading.
• Proofreading is a major cost and bottleneck in reconstructing neuronal networks.

Purpose of the Study:

• To develop a method for estimating connectomic proofreading progress.
• To provide a quantitative criterion for determining proofreading completeness.
• To reduce uncertainty in planning and budgeting proofreading efforts.

Main Methods:

• Utilizing global graph invariants that predictably converge to asymptotic limits.
• Estimating proofreading progress without prior knowledge of ground truth.
• Applying the method to zebrafish spinal cord and Drosophila melanogaster hemibrain datasets.

Main Results:

• Global graph invariants show predictable convergence patterns with increased proofreading.
• This convergence provides a quantitative 'pencils down' criterion for proofreading.
• The method effectively informs the completion status of connectomic reconstruction.

Conclusions:

• A fast, inexpensive method accurately estimates connectomic proofreading progress.
• This approach enables data owners to predict and budget necessary proofreading.
• The findings facilitate more efficient and reliable neuronal network reconstruction for scientific analysis.