EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

XMed-LLM: An Explainable Retrieval-Augmented Large Language Model Framework for Clinical Report Interpretation and Trustworthy Clinical Decision Support

Shweta Dharmadhikari, Bhagyashri Kulkarni, Ameya Kadam, Rupali Bhimrao Sirsat

Computer and Decision Making: An International Journal · 01.11.2026 · 3 (6S) · 2729-2636

DOI: 10.1922/comdem.v3i6s.19014 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 01.11.2026. Verfasst von Shweta Dharmadhikari und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Large language models (LLMs) are increasingly being explored for clinical report interpretation, medical question answering, summarization, and decision support. Their use in safety-critical settings, however, is constrained by hallucination, weak evidence traceability, limited interpretability, and uncertain clinician trust. Retrieval-augmented generation (RAG) can connect an LLM to external medical evidence, but retrieval alone does not guarantee that the retrieved material is relevant, correctly used, or transparently linked to the generated recommendation. Recent medical RAG studies report both gains and important failure modes, highlighting the need for stage-aware retrieval and evidence evaluation [1–5,7]. This paper proposes XMed-LLM, an explainable RAG framework for clinical report interpretation. The framework combines clinical report preprocessing, biomedical representation learning, hybrid lexical-semantic retrieval, evidence-grounded medical LLM inference, and a multi-layer explainability and trust module. The latter combines feature attribution (SHAP and LIME), attention visualization, explicit evidence citation, confidence estimation, and hallucination checks. The proposed evaluation plan uses MIMIC-III, MIMIC-IV and i2b2-type clinical resources and assesses retrieval, generation, predictive quality, explanation fidelity, evidence coverage, calibration, and clinician acceptance. Rather than treating explainability as a single post-hoc visualization, XMed-LLM links the retrieval trail and model output to a clinician-oriented explanation. The intended contribution is a practical architecture for more transparent, verifiable, and responsible generative AI-assisted clinical decision support.

Zitieren

APA

Dharmadhikari, S., Kulkarni, B., Kadam, A., & Sirsat, R. B. (2026). XMed-LLM: An Explainable Retrieval-Augmented Large Language Model Framework for Clinical Report Interpretation and Trustworthy Clinical Decision Support. Computer and Decision Making: An International Journal, 3(6S), 2729-2636. https://doi.org/10.1922/comdem.v3i6s.19014

Vancouver

Dharmadhikari S, Kulkarni B, Kadam A, Sirsat RB. XMed-LLM: An Explainable Retrieval-Augmented Large Language Model Framework for Clinical Report Interpretation and Trustworthy Clinical Decision Support. Computer and Decision Making: An International Journal. 2026;3(6S):2729-2636. doi:10.1922/comdem.v3i6s.19014

BibTeX
@article{Dharmadhikari2026,
  author = {Dharmadhikari, Shweta and Kulkarni, Bhagyashri and Kadam, Ameya and Sirsat, Rupali Bhimrao},
  title = {XMed-LLM: An Explainable Retrieval-Augmented Large Language Model Framework for Clinical Report Interpretation and Trustworthy Clinical Decision Support},
  journal = {Computer and Decision Making: An International Journal},
  year = {2026},
  volume = {3},
  number = {6S},
  pages = {2729--2636},
  doi = {10.1922/comdem.v3i6s.19014}
}
Kostenlos starten →