Multimodal Large Language Model-Enabled Healthcare Information System Design
International Journal of Healthcare Information Systems and Informatics · 2026-10-30 · 21 (1) · 1-22
DOI: 10.4018/ijhisi.426378 ↗Abstract
Cross-border health-care information systems often struggle to deliver culturally embedded medical knowledge—such as traditional Chinese medicine—to international users with varying health literacy, due to fragmented architecture and weak semantic alignment across languages. This study proposed a tripartite framework built on the Qwen2.5 Vision-Language model at the 7-billion parameter scale, integrating cross-lingual perception, domain knowledge enhancement, and terminology alignment to support multimodal health-care information processing. A bidirectional label-matching feedback mechanism and an entropy-weighted coupling coordination model were developed to quantify system-level synergy. Using 12,642 dissemination records and 7,869 health-care cases (2019–2025), the proposed system outperformed OpenAI's generative pretrained transformer 4o mini by 7.1 points on average across core indicators, with coupling coordination rising from 0.409 to 0.716 and regional synergy gains of 0.382 in Southeast Asia. The framework achieved a 39.2% retention rate in full-chain service conversion and over 83% adaptability for low-health-literacy users.
EUVIMED