Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Autoproof: Automated Segmentation Proofreading for Connectomics

Gary B Huang, William M Katz, Stuart Berg, Louis Scheffer

TL;DR

The paper tackles the proofreading bottleneck in EM connectomics by introducing Autoproof, an in-the-loop learning system trained on ground-truth proofreading data to automate focused-merge and orphan-link tasks. It validates the approach on a complete Drosophila CNS reconstruction, achieving $>90\%$ precision at $90\%$ recall for focused merges and recovering $90\%$ of the proofreading value at only $20\%$ of the effort, while automatically generating 200k merges in the orphan-link workflow and adding 309k T-bars and 2.4M PSDs to improve connectivity by $1.3$ percentage points. The method integrates a 3D CNN with multi-channel inputs, point-cloud shape representations, and synapse-level features, and demonstrates the potential for substantial time savings and scalable proofreading across large reconstructions. The authors discuss future directions including cross-dataset generalization, self-supervised learning, neurotransmitter signals, and novel imaging modalities to broaden applicability and robustness of automated proofreading in connectomics.

Abstract

Producing connectomes from electron microscopy (EM) images has historically required a great deal of human proofreading effort. This manual annotation cost is the current bottleneck in scaling EM connectomics, for example, in making larger connectome reconstructions feasible, or in enabling comparative connectomics where multiple related reconstructions are produced. In this work, we propose using the available ground-truth data generated by this manual annotation effort to learn a machine learning model to automate or optimize parts of the required proofreading workflows. We validate our approach on a recent complete reconstruction of the \emph{Drosophila} male central nervous system. We first show our method would allow for obtaining 90\% of the value of a guided proofreading workflow while reducing required cost by 80\%. We then demonstrate a second application for automatically merging many segmentation fragments to proofread neurons. Our system is able to automatically attach 200 thousand fragments, equivalent to four proofreader years of manual work, and increasing the connectivity completion rate of the connectome by 1.3\% points.

Autoproof: Automated Segmentation Proofreading for Connectomics

TL;DR

The paper tackles the proofreading bottleneck in EM connectomics by introducing Autoproof, an in-the-loop learning system trained on ground-truth proofreading data to automate focused-merge and orphan-link tasks. It validates the approach on a complete Drosophila CNS reconstruction, achieving precision at recall for focused merges and recovering of the proofreading value at only of the effort, while automatically generating 200k merges in the orphan-link workflow and adding 309k T-bars and 2.4M PSDs to improve connectivity by percentage points. The method integrates a 3D CNN with multi-channel inputs, point-cloud shape representations, and synapse-level features, and demonstrates the potential for substantial time savings and scalable proofreading across large reconstructions. The authors discuss future directions including cross-dataset generalization, self-supervised learning, neurotransmitter signals, and novel imaging modalities to broaden applicability and robustness of automated proofreading in connectomics.

Abstract

Producing connectomes from electron microscopy (EM) images has historically required a great deal of human proofreading effort. This manual annotation cost is the current bottleneck in scaling EM connectomics, for example, in making larger connectome reconstructions feasible, or in enabling comparative connectomics where multiple related reconstructions are produced. In this work, we propose using the available ground-truth data generated by this manual annotation effort to learn a machine learning model to automate or optimize parts of the required proofreading workflows. We validate our approach on a recent complete reconstruction of the \emph{Drosophila} male central nervous system. We first show our method would allow for obtaining 90\% of the value of a guided proofreading workflow while reducing required cost by 80\%. We then demonstrate a second application for automatically merging many segmentation fragments to proofread neurons. Our system is able to automatically attach 200 thousand fragments, equivalent to four proofreader years of manual work, and increasing the connectivity completion rate of the connectome by 1.3\% points.

Paper Structure

This paper contains 12 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Background
  3. Connectomics Segmentation Workflows
  4. Related Work
  5. Methods
  6. CNN for Focused Merge
  7. Additional Information
  8. Results
  9. Data
  10. Focused Merge
  11. Orphan Link
  12. Conclusions and Future Work

Figures (2)

  • Figure 1: Focused merge precision-recall. Baseline performance using scores from the initial segmentation algorithm is in blue (baseline agglo). Performance using the CNN is given in orange (convnet), and performance using the CNN combined with shape and synapse information is give in green (convnet++).
  • Figure 2: Precision of randomly sampled Autoproof proposed orphan link merges. Dotted curves show precision if all merges marked "indeterminate" are considered false, solid curves if considered true. Green combined curve shows precision if a merge is considered true if either proofreader A or B marked the merge as true.