03 EM Prep and Imaging

Session outcomes (60 minutes)

• Identify major artifact classes and their downstream failure modes.
• Define QA gates before large-scale reconstruction.
• Build a practical acquisition risk register with mitigation triggers.

Pedagogical arc

• Hook: artifact examples and scientific consequences.
• Model: prep chain + QA checkpoints.
• Practice: classify artifacts and set gate thresholds.
• Check: triage decisions under constrained throughput.

Acquisition chain (operational)

Fixation -> Staining -> Sectioning/block-face -> Imaging -> Stack assembly

Each stage introduces distinct, diagnosable error signatures.

Real data: one section of human cortex

• H01: a ~4 mm wedge, 5,019 sections at 33.9 nm mean thickness.
• 1.8 PB raw acquisition; 326 days of 61-beam microscope time.

Real data: the resolution you must reach everywhere

• Same sample at native 4 nm: membranes, mitochondria, vesicle clouds.
• Synapse-finding needs this resolution across the whole 4 mm volume.

Artifact taxonomy with downstream impact

• Physical (tear/fold/chatter): topology discontinuities.
• Signal (charging/contrast drift): false boundaries, missed synapses.
• Geometric (misalignment/seams): apparent neurite breaks/merges.

QA gates before full ingest

• Signal stability: SNR and intensity drift bounds.
• Geometric consistency: seam residual and alignment error limits.
• Completeness: missing/damaged section accounting.
• Metadata completeness: acquisition parameters and provenance.

Pilot-first strategy

• Run pilot segmentation/QC on representative blocks.
• Quantify expected merge/split burden before full-volume processing.
• Adjust prep protocol if projected correction load is unacceptable.

Throughput vs fidelity tradeoff

• Higher throughput without gates amplifies downstream correction cost.
• QA cost upfront is often cheaper than post hoc proofreading.
• Teach teams to model this explicitly, not intuitively.

Misconceptions to correct

• "Artifacts can be cleaned up later without scientific cost."
• "Good visual quality implies quantitative adequacy."
• "Metadata can be reconstructed after acquisition."

Activity: risk register workshop

For three artifact types, specify:

• detection metric,
• alert threshold,
• mitigation action,
• stop/go decision owner.

Formative rubric

• Pass: each artifact has metric + threshold + mitigation.
• Strong: downstream impact quantified in scientific terms.
• Flag: qualitative descriptions without operational triggers.

External paper figure slots

• Hayworth et al. EM imaging methodology figures (prep-to-imaging pipeline).
• Large-scale EM platform figure (multibeam / acquisition throughput tradeoff).
• QA dashboard example from open connectomics workflow publications.

Bridge

Next unit: infrastructure for robust reconstruction once acquisition is trusted.