The Journal of craniofacial surgery · 2026
Active Knowledge Retrieval to Reduce Hallucinations and Enhance Factual Accuracy in Large Medical Reasoning Models.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
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.
Lian Heqing, Wu Jingwei, Wang Zeyu, Chen Lei, Ji Lei, Ren Lina
Аннотация
OBJECTIVE: Large reasoning models (LRMs) have demonstrated robust performance in closed-domain tasks; however, their limitations persist in medical applications that require high factual precision and a strong reliance on dynamic, domain-specific knowledge.
METHOD: This study introduces a reasoning-retrieval-information integration framework designed to enable LRMs to autonomously access real-time medical data during inference. By combining their inherent model knowledge with externally retrieved, up-to-date medical data, the framework emulates the diagnostic and decision-making processes commonly used in clinical practice. The proposed method addresses key limitations of conventional retrieval-augmented generation and agent-based approaches, particularly their constrained integration with LRMs.
RESULT: Significant performance improvements were observed in open-domain medical question-answering tasks. Using sleep medicine as a representative case, the framework was evaluated across several core clinical functions, including disease diagnosis, etiology analysis, treatment planning, medical recommendation, outcome interpretation, condition description, prognosis communication, and patient counseling. Results demonstrated superior performance across multiple evaluation metrics compared with existing methodologies.
CONCLUSION: These findings support the development of more interpretable and clinically applicable medical AI systems, and offer a scalable foundation for knowledge-intensive reasoning tasks in other domains.
METHOD: This study introduces a reasoning-retrieval-information integration framework designed to enable LRMs to autonomously access real-time medical data during inference. By combining their inherent model knowledge with externally retrieved, up-to-date medical data, the framework emulates the diagnostic and decision-making processes commonly used in clinical practice. The proposed method addresses key limitations of conventional retrieval-augmented generation and agent-based approaches, particularly their constrained integration with LRMs.
RESULT: Significant performance improvements were observed in open-domain medical question-answering tasks. Using sleep medicine as a representative case, the framework was evaluated across several core clinical functions, including disease diagnosis, etiology analysis, treatment planning, medical recommendation, outcome interpretation, condition description, prognosis communication, and patient counseling. Results demonstrated superior performance across multiple evaluation metrics compared with existing methodologies.
CONCLUSION: These findings support the development of more interpretable and clinically applicable medical AI systems, and offer a scalable foundation for knowledge-intensive reasoning tasks in other domains.