EUVIMEDEUVIMED entdecken
Crossref

Knowledge Graph-guided Inference for Medical VQA: Enabling Lightweight and Diverse Clinical Applications

Rintaro Nagami, Tomoki Hamagami

Journal of Information Processing · 2026 · 34 · 344-354

DOI: 10.2197/ipsjjip.34.344 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Information Processing, 2026. Verfasst von Rintaro Nagami und 1 weiteren Autor:innen. Die Arbeit nennt 20 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study aims to develop a lightweight yet accurate Medical Visual Question Answering (MedVQA) model to support automatic report generation from diagnostic images. While large-scale language models (LLMs) have demonstrated impressive performance in medical text generation, their high computational demands pose challenges for practical deployment in clinical settings. To address this, we propose novel approaches that enable small-scale models to achieve high performance by leveraging structured domain knowledge. We introduce two graph-based methods that hierarchically integrate background knowledge via knowledge graphs. The first method enhances contextual reasoning by incorporating causal and domain-specific relations; while the second method fuses image and knowledge graphs to capture multimodal dependencies. Experimental results show that Method 1 improves accuracy on logic-intensive datasets such as Path-VQA and PMC-VQA, achieving performance comparable to models with approximately seven times more parameters. Method 2 proves particularly effective in handling abstract and spatial question types in the VQA-RAD dataset. Furthermore, feature space visualizations reveal that high-performing models form broader and semantically richer feature distributions even within the same question type, suggesting increased representational flexibility. These findings indicate that our graph-based approach enhances the reasoning capabilities of compact models, offering a scalable and explainable solution suitable for clinical use.

Zitieren

APA

Nagami, R., & Hamagami, T. (2026). Knowledge Graph-guided Inference for Medical VQA: Enabling Lightweight and Diverse Clinical Applications. Journal of Information Processing, 34(0), 344-354. https://doi.org/10.2197/ipsjjip.34.344

Vancouver

Nagami R, Hamagami T. Knowledge Graph-guided Inference for Medical VQA: Enabling Lightweight and Diverse Clinical Applications. Journal of Information Processing. 2026;34(0):344-354. doi:10.2197/ipsjjip.34.344

BibTeX
@article{Nagami2026,
  author = {Nagami, Rintaro and Hamagami, Tomoki},
  title = {Knowledge Graph-guided Inference for Medical VQA: Enabling Lightweight and Diverse Clinical Applications},
  journal = {Journal of Information Processing},
  year = {2026},
  volume = {34},
  number = {0},
  pages = {344--354},
  doi = {10.2197/ipsjjip.34.344}
}
Kostenlos starten →