Unit: Volume Reconstruction Infrastructure

Learning goals

• Describe the end-to-end infrastructure required to reconstruct large EM volumes.
• Explain key system components: ingest, storage, preprocessing, and service APIs.
• Identify scaling bottlenecks and reliability tradeoffs in reconstruction pipelines.

Scope boundaries

• In scope:
  • systems architecture for large-scale connectomics reconstruction
  • data movement, preprocessing, and service orchestration
• Out of scope:
  • low-level imaging acquisition details (Unit 03)
  • advanced analysis/NeuroAI interpretation (Unit 09)

Website draft blocks

Hero framing

Reconstructing connectomes is a systems problem as much as a neuroscience problem. This unit covers the infrastructure patterns that make petascale-to-exascale reconstruction feasible.

Core architecture layers

• Ingest and normalization services.
• Multi-resolution storage and access patterns.
• Preprocessing and transformation pipelines.
• API/service layer for downstream tools and users.

Operational concerns

• Throughput vs accuracy tradeoffs.
• Failure handling and observability.
• Cost/performance constraints at scale.

Why this matters

Infrastructure choices shape what analyses are feasible, how reproducible results are, and how quickly scientific teams can iterate.

Slide draft sequence (v1)

1. Why infrastructure is the hidden bottleneck
2. End-to-end reconstruction architecture
3. Ingest and normalization
4. Storage/multiscale representation
5. Preprocessing pipeline stages
6. Service/API layer
7. Throughput, cost, and reliability tradeoffs
8. Case study: scaling lessons learned
9. Integration with segmentation/proofreading and analysis units
10. Summary + architecture checklist

Figure candidates

See: course/units/figures/04-volume-reconstruction-infrastructure-selected-v1.md

Cross-links

• Related modules: module12, module18
• Related units: 03-em-prep-and-imaging, 08-segmentation-and-proofreading
• Related datasets: /datasets/workflow/, /datasets/mouseconnects/

Open issues

• Add a simplified reference architecture diagram for web readability.
• Separate “historical implementation” from “recommended current pattern” language.

Scope check (expert pass)

• Cover alignment, stitching, normalization, and chunked storage explicitly.
• Include provenance/versioning requirements for derived volumes and annotations.
• Treat workflow orchestration and failure recovery as core, not optional, system concerns.

Technical anchors to preserve

• Multi-resolution pyramids for interactive and batch workloads.
• Separation of storage, compute, and annotation services.
• Cost-aware architecture choices for petascale workloads.