Specialty-Specific Citation-Enabled AI Clinical Decision Support System for Craniofacial Surgery: Development of...

Area of Science:

• Medical AI
• Surgical Decision Support
• Craniofacial Surgery

Background:

• Craniofacial surgery demands complex, multidisciplinary knowledge integration.
• Current decision support tools are specialty-specific and lacking.
• Retrieval-augmented generation (RAG) presents an opportunity for AI in surgical practice.

Purpose of the Study:

• To develop and evaluate CASPER, a domain-specific, multimodal RAG system for craniofacial surgery.
• To assess the system's ability to retrieve and synthesize evidence-based information for clinical questions.
• To validate the system's performance against expert-level reasoning and literature.

Main Methods:

• Developed CASPER, a multimodal RAG system using RAPTOR architecture.
• Utilized a knowledge base of 8561 craniofacial surgery articles (2000-2025).
• Evaluated performance via semantic similarity, content coverage, and citation accuracy for 25 clinical questions.

Main Results:

• CASPER demonstrated strong literature alignment (mean SEM-eval 0.89).
• Achieved high performance in pediatric airway, facial trauma, and oncologic surgery.
• Manual review confirmed high content coverage and accurate citations.

Conclusions:

• CASPER is the first citation-enabled AI for craniofacial surgery, showing expert-comparable reasoning.
• The system offers transparent, evidence-grounded recommendations for surgical planning.
• CASPER can enhance decision-making consistency and knowledge translation in craniofacial surgery education and practice.