Unit: Segmentation and Proofreading

Learning goals

• Explain why segmentation quality is a limiting factor in connectomics.
• Identify common segmentation/proofreading error classes in EM volumes.
• Apply a practical proofreading workflow for correction and verification.
• Connect proofreading outcomes to downstream graph and analysis reliability.

Scope boundaries

• In scope:
  • segmentation outputs and human QC workflows
  • merge/split/boundary ambiguity error taxonomy
  • practical morphology cues from mature proofreading material
• Out of scope:
  • full wet-lab EM acquisition pipeline
  • advanced NeuroAI motif-analysis theory (Unit 09)

Source synthesis

• Core claims to preserve:
  • proofreading is essential because segmentation errors propagate to scientific inference
  • morphology-informed classification (axon/dendrite/ultrastructure/glia cues) improves QC accuracy
  • quality metrics should be attached to correction workflows, not added as afterthoughts
• Historical context to reframe:
  • older outreach pipeline visuals are context-only, not current-state authority
• Needs verification:
  • any performance or benchmark claims from outreach-derived slides

Website draft blocks

Hero framing

Segmentation makes connectome-scale analysis possible, but proofreading makes it reliable. This unit teaches how to recognize reconstruction errors, correct them systematically, and document quality in ways that protect downstream scientific conclusions.

Error taxonomy

• Merge errors: two processes incorrectly fused.
• Split errors: one process incorrectly fragmented.
• Boundary ambiguity: uncertain edges in dense neuropil.
• Identity confusion: neuronal/glial or axon/dendrite misclassification.

Practical proofreading workflow

1. Inspect local morphology and continuity in 2D/3D context.
2. Classify suspected error type.
3. Correct labels/links using tooling workflow.
4. Verify correction across neighboring slices and context views.
5. Record QC notes and confidence for auditability.

Morphology cues used in QC

• Dendritic spine and shaft cues.
• Axonal bouton and process continuity cues.
• Synaptic vesicle/ultrastructure cues.
• Glia-neuron boundary cues in ambiguous regions.

QC and downstream impact

• Error type distributions are quality signals.
• Corrections affect graph topology and inferred motifs.
• Reliable proofreading supports valid connectome analysis and NeuroAI conclusions.

Slide draft sequence (v1)

1. Why proofreading matters for connectomics validity
2. Segmentation output and failure modes
3. Error taxonomy: merges, splits, boundary ambiguity
4. Axon/dendrite confusion cases
5. Ultrastructure cues for correction decisions
6. Glia-vs-neuron boundary pitfalls
7. Step-by-step proofreading workflow
8. Tool-assisted correction and logging
9. QC metrics and acceptance thresholds
10. Case study: before/after corrections
11. How QC affects downstream graph analysis
12. Summary + practical checklist

Figure candidates

See: course/units/figures/08-segmentation-and-proofreading-selected-v1.md

Cross-links

• Related modules: module06, module07
• Related datasets: /datasets/mouseconnects/, /datasets/workflow/
• Related tools: /tools/connectome-quality/, /tools/ask-an-expert/

Open issues

• Add one notebook/lab exercise path for hands-on proofreading.
• Define rubric thresholds for student-level proofreading assignments.
• Validate any outreach-derived metric claims before publish.

Scope check (expert pass)

• Add metrics vocabulary beyond generic QC: variation of information (VI), edge-level precision/recall, expected run length (ERL), synapse-centric F1.
• Require explicit pre/post correction logging for reproducibility.
• Include error-priority policy: high-impact topology/synapse errors first.

Technical anchors to preserve

• Proofreading is a scientific quality-control process, not only visual cleanup.
• Metrics and correction workflow must be linked.
• Human-machine workflows should separate discovery, adjudication, and finalization steps.