MedScreenDental: Automated structured dental record generation via multimodal language model integration

Open Webpage

Wenzhong Jin, Yilan Sun, Kaiyuan Ji, Xiaoyan Jiang, Yufeng Hu, Jinwu Wang, Jiannan Liu

Published: 01 Jan 2025, Last Modified: 13 Nov 2025Displays 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Developed MedScreenDental—a non-invasive LLM-OCR framework for outpatient documentation structuring, enabling cross-platform EHR integration without API modifications via screen-capture processing and adaptive multimodal LLM workflows.•The system enables dynamic model scheduling via real-time request-state analysis and expandable MCP Servers, supporting seamless integration of new LLMs while maintaining clinical workflow compatibility.•The multimodal model Qwen2.5-VL-72B-Instruct shows the best overall performance, with OCR+DeepSeek-V3 excelling in the Past Medical History domain, achieving state-of-the-art results in symptom extraction and clinical reasoning.•Established the first oral disease benchmark (1,014 cases: periodontal/mucositis/tumors) with hybrid metrics BLEU/ROUGE.