[3F1-OS-8] OS-8

10:00 AM - 10:15 AM JST(1:00 AM - 1:15 AM UTC)

[3F1-OS-8-05] An RLM-Based Approach to Automated Iterative Annotation and Quality Control of Case Report Texts

〇Eisuke Dohi¹, Jin-Dong Kim², Itaru Hayakawa³, Tomoyasu Matsubara⁴, Terue Takatsuki², Yuka Tateishi², Toyofumi Fujiwara², Yasunori Yamamoto² (1. National Center of Neurology and Psychiatry, 2. Research Organization of Information and Systems, Database Center for Life Science, 3. National Center for Child Health and Development, Department of Neurology, 4. Hiroshima University, Department of Neurology)

Keywords:

RLM,Automatic Annotation,Case Report,Iterative Processing,Quality Control

We propose an RLM (Recursive Language Models)-based iterative annotation system for medical case report texts. Unlike single-pass LLM approaches, our method validates extraction results and performs additional annotation when predefined quality criteria are not satisfied. We introduce Event Coverage, a clinical-event-based evaluation metric, and dynamically switch strategies according to real-time quality indicators. In preliminary experiments, iterative refinement improved Event Coverage from 54% (single-pass) to 96% with an average of two iterations. The system was developed in collaboration with practicing clinicians, reflecting practical requirements for high-quality pre-review drafts. Our results suggest that iterative quality control can enhance completeness while enabling local LLM deployment for privacy-preserving clinical applications.