Defect Annotation TASK-05
Marking imaging artifacts, tissue damage, and data quality issues.
Task Overview
What is Defect Annotation?
Defect annotation is the systematic identification and labeling of regions where imaging or sample preparation problems compromise data quality. This includes tissue tears, folds, missing sections, staining artifacts, and imaging distortions. Marking these areas prevents wasted effort on unannotatable regions.
Why is it Important?
Defects are inevitable in large-scale EM datasets. Without proper annotation, proofreaders waste time on impossible-to-resolve regions, and AI models may learn incorrect patterns from corrupted data. Defect masks enable automated quality metrics and guide human attention to regions that can actually be improved.
Key Objectives
- Identify all major defect types in assigned regions
- Create accurate boundary masks for affected areas
- Classify defects by type and severity
- Document extent of data loss for quality reporting
- Flag regions requiring re-imaging if possible
Current Standard Operating Procedure Authoritative
Defect Classification
Mark each defect with the appropriate category:
- Section Loss — Missing tissue section (gap in Z)
- Tissue Fold — Section folded over itself
- Tissue Tear — Rip through tissue
- Staining Artifact — Over/under staining, precipitate
- Knife Mark — Cutting damage (chatter, compression)
- Imaging Artifact — Focus issues, charging, drift
- Contamination — Foreign material on section
Annotation Procedure
- Load the assigned volume in WebKnossos
- Systematically scan through all Z-slices
- When a defect is found, select the appropriate defect label
- Paint the affected region, including a small margin
- Ensure the mask covers all unusable tissue
- Continue scanning through the entire volume
- Submit with notes on any unusual findings
Severity Levels
- Minor — Annotation possible with effort
- Moderate — Some structures recoverable
- Severe — Region completely unusable
Quality Criteria
- All visible defects identified and labeled
- Defect boundaries accurate to within 1 µm
- Correct classification applied to each defect
- No over-annotation of usable tissue
Historical SOPs Archived
Previous defect annotation protocols are preserved here for reference. Always use the Current SOP above.
Version History
-
v1.0 (Current) — 2025
Initial defect classification system
Visual Examples
Reference images showing each defect type and correct annotation boundaries.
Defect Visual Reference
- Section Loss — Abrupt change in image content between Z-slices
- Tissue Fold — Double-layer appearance, overlapping structures
- Tear — Dark void with ragged edges
- Staining — Unusually light or dark regions, crystalline deposits
- Knife Mark — Parallel streaks across section
Common Failure Modes
- Missing subtle defects Some artifacts are hard to see at low zoom. Always review at full resolution when scanning for defects.
- Over-annotating as defect Not all unusual-looking tissue is defective. Dense neuropil and myelin are often misidentified as staining artifacts.
- Imprecise boundaries Defect masks should follow actual damage, not approximate it. Overly large masks waste usable data.
- Missing multi-section defects Some defects span many sections (deep tears, folds). Track them through Z to ensure complete coverage.
- Wrong classification Correct classification helps prioritize fixes. A knife mark requires different remediation than a staining issue.
Tools Used
WebKnossos
Primary defect annotation platform
Neuroglancer
Large-scale visualization for survey
Annotation Tools
- Brush — Paint defect regions
- Eraser — Remove over-annotation
- Fill — Flood-fill enclosed regions
- Interpolation — Extend masks across Z-slices