Neurosurgical Book Parser Skill

What This Skill Does

Extracts structured medical knowledge from neurosurgical and spine surgery textbooks:

• Anatomical structures and their spatial relationships
• Surgical procedures with step-by-step sequences
• Clipping/fixation techniques and their applications
• Complications and contraindications
• Clinical outcomes and evidence

When to Use This Skill

Activate when the user:

• Wants to process a neurosurgical or spine surgery textbook
• Asks to build a medical knowledge graph
• Needs to extract surgical procedures from PDFs
• Wants to create clinical decision support content
• Mentions "Seven Aneurysms", "spine surgery", "aneurysm clipping", or similar

Critical Workflow: Direct Reading Over API

ALWAYS read the book content directly rather than using API calls.

See docs/extraction-lessons-learned.md for why:

• API extraction: $50-100 per book, 26+ hours, unreliable
• Direct reading: $0, ~1 hour, 100% reliable

Extraction Workflow

Step 1: Identify Book Structure

First, map the book's organization:

# Check for MinerU-processed content
ls mineru_output/*/auto/*.md

# Read table of contents or first pages
head -200 mineru_output/*/auto/*.md | grep -E "^#|Chapter|Section"

Create a chapter-to-page mapping like:

CHAPTER_PAGES = {
    "Ch1_Introduction": (1, 15),
    "Ch2_Anatomy": (16, 45),
    # ... map all chapters
}

Step 2: Define Entity Types

Use domain-specific types from ENTITY-TAXONOMY.md:

Neurosurgical:

• artery, vein, nerve, cistern, brain_region, bone_structure
• surgical_step, clipping_technique, surgical_approach
• aneurysm, complication, instrument

Spine:

• vertebra, disc, nerve_root, ligament, foramen
• decompression_technique, fusion_technique, fixation_approach
• stenosis, herniation, myelopathy, screw_type

Step 3: Extract Entities by Chapter

Read each chapter and extract entities with context:

# Read a chapter section
# Note which chapter/section/page for context

For each entity, capture:

• name: Canonical name (lowercase, specific)
• entity_type: From ENTITY-TAXONOMY.md
• page: Source page number
• chapter: Chapter reference
• description: Brief context from text

Step 4: Insert into Neo4j

Use batched insertions (20-30 at a time):

docker exec neurosurgery-neo4j cypher-shell -u neo4j -p "neo4j_dev_pass_2025" "
CREATE (:Anatomy {name: 'middle cerebral artery', book_entity_type: 'artery', page: 79, chapter: 'Ch15_MCA'})
CREATE (:Anatomy {name: 'm1 segment', book_entity_type: 'artery', page: 79, chapter: 'Ch15_MCA'})
// ... more entities
"

Step 5: Build Relationships

Connect entities with surgical knowledge flow:

docker exec neurosurgery-neo4j cypher-shell -u neo4j -p "neo4j_dev_pass_2025" "
MATCH (a:Pathology {name: 'mca aneurysm'}), (c:Anatomy {name: 'sylvian cistern'})
CREATE (a)-[:LOCATED_IN {context: 'MCA aneurysms in sylvian cistern'}]->(c)
"

Step 6: Log in KAIZEN.md

After each extraction, update KAIZEN.md with:

• Book processed
• Entity/relationship counts
• New patterns discovered
• Mistakes avoided

Neo4j Label Mapping

Entity Type	Neo4j Label
artery, vein, nerve, cistern, brain_region	Anatomy
surgical_step	Procedure
clipping_technique, fusion_technique	Technique
aneurysm, stenosis, complication	Pathology
instrument, screw_type	Instrument
figure, chapter, tenet	Reference

Relationship Types

Anatomical:

• LOCATED_AT, ADJACENT_TO, SUPPLIES, DRAINS_TO
• PASSES_THROUGH, BRANCHES_FROM

Surgical:

• REQUIRES_STEP, FOLLOWED_BY, USES_TECHNIQUE
• USES_INSTRUMENT, APPLIES_TO, TREATS, PROVIDES_ACCESS_TO

Knowledge:

• ILLUSTRATED_BY, DESCRIBED_IN, WARNS_ABOUT, REFERENCED_IN

Clinical:

• COMPLICATES, INDICATES, CONTRAINDICATES

Supporting Files

• ENTITY-TAXONOMY.md - Complete entity type definitions
• EXTRACTION-PATTERNS.md - Proven extraction patterns
• COMMON-MISTAKES.md - Anti-patterns to avoid
• KAIZEN.md - Continuous improvement log
• examples/ - Case studies from prior extractions

Integration with Project

This skill integrates with:

• neurosurgery_db/ingestion/parser.py - MinerU JSON parsing
• neurosurgery_db/ingestion/graph_loader.py - Neo4j graph loading
• docker-compose.yml - Neo4j container configuration

Quality Checklist

Before finishing an extraction:

• All chapters processed
• Entity names are canonical (lowercase, specific)
• Page references preserved
• Surgical steps are sequenced (FOLLOWED_BY)
• Aneurysms/pathologies linked to approaches
• Complications linked to procedures
• KAIZEN.md updated with learnings