EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Medical LLMs: Fine-Tuning vs. Retrieval-Augmented Generation

Bhagyajit Pingua iD, Adyakanta Sahoo iD, Meenakshi Kandpal iD, Deepak Murmu iD, Jyotirmayee Rautaray iD, Rabindra Kumar Barik iD, Manob Jyoti Saikia iD

Bioengineering · 24.06.2025 · 12 (7) · 687

DOI: 10.3390/bioengineering12070687 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Bioengineering, online erschienen am 24.06.2025. Verfasst von Bhagyajit Pingua und 6 weiteren Autor:innen. Beteiligte Einrichtung: Biomedical Sensors & Systems Lab, University of Memphis, Memphis, TN 38152, USA. Die Arbeit wurde bisher 38-mal zitiert (Crossref) und nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Biomedical Sensors & Systems Lab, University of Memphis.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Large language models (LLMs) are trained on huge datasets, which allow them to answer questions from various domains. However, their expertise is confined to the data that they were trained on. In order to specialize LLMs in niche domains like healthcare, various training methods can be employed. Two of these commonly known approaches are retrieval-augmented Generation and model fine-tuning. Five models—Llama-3.1-8B, Gemma-2-9B, Mistral-7B-Instruct, Qwen2.5-7B, and Phi-3.5-Mini-Instruct—were fine-tuned on healthcare data. These models were trained using three distinct approaches: retrieval-augmented generation (RAG) alone, fine-tuning (FT) alone, and a combination of both (FT+RAG) on the MedQuAD dataset, which covers a wide range of medical topics including disease symptoms, treatments, medications, and more. Our findings revealed that RAG and FT+RAG consistently outperformed FT alone across most models, particularly LLAMA and PHI. LLAMA and PHI excelled across multiple metrics, with LLAMA showing superior overall performance and PHI demonstrating strong RAG/FT+RAG capabilities. QWEN lagged behind in most metrics, while GEMMA and MISTRAL showed mixed results.

Schlagwörter

Stichwörter der Autor:innen

fine-tuninghealthcarelarge language modelsmedicalretrieval-augmented generation

Autor:innen und Institutionen

  • Biomedical Sensors & Systems Lab, University of Memphis, Memphis, TN 38152, USA.
  • Department of Computer Science and Engineering, Odisha University of Technology and Research, Bhubaneswar 751003, India.
  • Department of Information Technology, Veer Surendra Sai University of Technology, Sambalpur 768018, India.
  • School of Computer Applications, KIIT Deemed to be University, Bhubaneswar 751003, India.
  • Electrical and Computer Engineering Department, University of Memphis, Memphis, TN 38152, USA.

ORCID-Profile: Bhagyajit Pingua, Adyakanta Sahoo, Meenakshi Kandpal, Deepak Murmu, Jyotirmayee Rautaray, Rabindra Kumar Barik, Manob Jyoti Saikia

Förderung

  • Biomedical Sensors & Systems Lab, University of Memphis

Zitieren

APA

Pingua, B., Sahoo, A., Kandpal, M., Murmu, D., Rautaray, J., Barik, R. K., & Saikia, M. J. (2025). Medical LLMs: Fine-Tuning vs. Retrieval-Augmented Generation. Bioengineering, 12(7), 687. https://doi.org/10.3390/bioengineering12070687

Vancouver

Pingua B, Sahoo A, Kandpal M, Murmu D, Rautaray J, Barik RK, et al. Medical LLMs: Fine-Tuning vs. Retrieval-Augmented Generation. Bioengineering. 2025;12(7):687. doi:10.3390/bioengineering12070687

BibTeX
@article{Pingua2025,
  author = {Pingua, Bhagyajit and Sahoo, Adyakanta and Kandpal, Meenakshi and Murmu, Deepak and Rautaray, Jyotirmayee and Barik, Rabindra Kumar and Saikia, Manob Jyoti},
  title = {Medical LLMs: Fine-Tuning vs. Retrieval-Augmented Generation},
  journal = {Bioengineering},
  year = {2025},
  volume = {12},
  number = {7},
  pages = {687},
  doi = {10.3390/bioengineering12070687}
}
Kostenlos starten →