DOI 10.1097/scs.0000000000012279
Specialty-Specific Citation-Enabled AI Clinical Decision Support System for Craniofacial Surgery: Development of CASPER.
Аннотация
BACKGROUND
Craniofacial surgery requires synthesis of complex, multidisciplinary knowledge, yet specialty-specific decision support tools are lacking. Retrieval-augmented generation (RAG) offers an opportunity to create transparent, evidence-based artificial intelligence (AI) systems tailored to surgical practice.
METHODS
The authors developed CASPER, a domain-specific, multimodal RAG system with text and image analysis capabilities, built with RAPTOR hierarchical architecture and a knowledge base of 8561 open-access craniofacial surgery articles (2000-2025). The system retrieved and synthesized peer-reviewed literature in response to 25 clinical questions spanning craniofacial subspecialties. Performance was evaluated using semantic similarity (SEM-eval) to retrieved documents, manual content coverage review, and manual citation accuracy verification.
RESULTS
CASPER achieved strong alignment with supporting literature (mean SEM-eval 0.89±0.04; range: 0.81-0.96), integrating an average of 7.8 sources per query. Highest performance was observed in pediatric airway (0.93), facial trauma (0.93), and oncologic surgery (0.95), with top scores for Pierre-Robin mandibular distraction (0.96) and orbital floor fracture management (0.95). Lower scores occurred in complex or emerging domains such as Le Fort III contraindications (0.81) and facial feminization planning (0.81). Manual review confirmed that CASPER maintained high content coverage across scenarios, with citations consistently accurate and directly supportive of system outputs.
CONCLUSIONS
CASPER is the first citation-enabled AI system for craniofacial surgery and demonstrates expert-comparable reasoning across diverse clinical scenarios. By delivering transparent, evidence-grounded recommendations, CASPER has the potential to enhance surgical planning, improve decision-making consistency, and accelerate knowledge translation in both clinical practice and surgical education.
Предметные рубрики MeSH
Подготовка и перевод публикации
Обработка статьи · этапы, ошибки и управление
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Тип публикации указан по данным источника. Уровень доказательности автоматически не присваивается.