Skip to content
DraconiticResearch
Search Sources API
203,234 studies31 sourcesUpdated 43 min ago
Peer-reviewedCC BY 4.0Japanese Journal of Radiology

Open-source offline-deployable retrieval-augmented large language model for assisting pancreatic cancer staging

  • Hisashi Johno1
  • Akitomo Amakawa1
  • Atsushi Komaba1
  • Ryota Tozuka2,3
  • Yuki Johno4
  • Junichi Sato1
  • Kentaro Yoshimura5
  • Kazunori Nakamoto6
  • Shintaro Ichikawa1
6 affiliations
  1. 1Department of Diagnostic Radiology, Faculty of Medicine, University of Yamanashi, 1110 Shimokato, Chuo, Yamanashi Japan
  2. 2Department of Therapeutic Radiology, Faculty of Medicine, University of Yamanashi, Yamanashi, Japan
  3. 3Department of Radiation Oncology, Tohoku University Graduate School of Medicine, Sendai, Japan
  4. 4Department of Internal Medicine, Kyonan Medical Center Fujikawa Hospital, Yamanashi, Japan
  5. 5Division of Molecular Biology, Center for Medical Education and Sciences, Faculty of Medicine, University of Yamanashi, Yamanashi, Japan
  6. 6Division of Mathematics, Center for Medical Education and Sciences, Faculty of Medicine, University of Yamanashi, Yamanashi, Japan
Read full textPDFDOICite

Abstract

Purpose: Large language models (LLMs) are increasingly applied in radiology, but key challenges remain, including data leakage from cloud-based systems, false outputs, and limited reasoning transparency. This study aimed to develop an open-source, offline-deployable retrieval-augmented LLM (RA-LLM) system in which local execution prevents data leakage and retrieval-augmented generation (RAG) improves output accuracy and transparency using reliable external knowledge (REK), demonstrated in pancreatic cancer staging.

Materials and methods: Llama-3.2 11B and Gemma-3 27B were used as local LLMs, and GPT-4o mini served as a cloud-based comparator. The Japanese pancreatic cancer guideline, written in English, served as REK. Relevant REK excerpts were retrieved to generate retrieval-augmented responses. System performance, including classification accuracy, retrieval metrics, and execution time, was evaluated on 100 simulated pancreatic cancer cases with CT findings described in English, using non-RAG LLMs as baselines. McNemar tests were applied to TNM staging and resectability classification.

Results: RAG improved TNM staging accuracy for all LLMs (GPT-4o mini 61% → 90%, p < 0.001; Llama-3.2 11B 53% → 72%, p < 0.001; Gemma-3 27B 59% → 87%, p < 0.001) and mildly improved resectability classification for GPT-4o mini and Llama-3.2 11B (72% → 84%, p = 0.012; 58% → 73%, p = 0.006), while the improvement for Gemma-3 27B was not evident (77% → 86%, p = 0.093). Gemma-3 27B showed performance comparable to GPT-4o mini. Retrieval performance was high (context recall = 1; context precision = 0.5–1), and local models ran at speeds comparable to the cloud-based GPT-4o mini.

Conclusion: We developed an offline-deployable RA-LLM system for pancreatic cancer staging and publicly released its full source code. RA-LLMs outperformed baseline LLMs, and the offline-capable Gemma-3 27B performed comparably to the widely used cloud-based GPT-4o mini.

Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1007/s11604-026-02049-8.

Related

  1. AFP assistant: a retrieval-augmented generation and large language model-powered multilingual polio chatbot for low-resource language communities
    Gelane Biru, Honey Gemechu, Firanol Teshome +8
    npj Digital MedicinePeer-reviewedCC BYFull text1 citedEpidemiology & public healthDOAJ
  2. Evidence-based clinical question answering using retrieval-augmented generation (RAG) and open-source large language models
    Ayesha Qasim, Jesus Jaime Moreno Escobar, Christian E. Maldonado-Sifuentes +3
    Scientific ReportsPeer-reviewedCC BY 4.0Full textEpidemiology & public healthCrossref
  3. HalluciGuard-eye: knowledge graph-enhanced retrieval-augmented generation for hallucination detection and mitigation in ophthalmic multimodal large language models
    Ge Wu, Chong Ai, Weiqi Wu
    Frontiers in EndocrinologyPeer-reviewedCC BY 4.0Full textEndocrinology & metabolismCrossrefEurope PMCPMC
  4. Medical Students' Perceptions of Retrieval-Augmented Generation and Large Language Models: Dyadic Interview and Exploratory Survey Study.
    Rohin Athavale, Alexander Cresswell, Alice Huffman
    JMIR formative researchPeer-reviewedCC BYFull textEpidemiology & public healthEurope PMCPMCDOAJ
  5. BanglaDiaHallu: Hallucinations of large language models and the effect of guideline-grounded retrieval-augmented generation in Bangla diabetes question answering
    K. Nahid, M. Anam, M. T. H. Saon +3
    medRxivPreprintCC BYFull textEpidemiology & public healthmedRxiv

Cite

BibTeXRIS
BibTeX
@article{johno2026opensource,
  title = {{Open-source offline-deployable retrieval-augmented large language model for assisting pancreatic cancer staging}},
  author = {Hisashi Johno and Akitomo Amakawa and Atsushi Komaba and Ryota Tozuka and Yuki Johno and Junichi Sato and Kentaro Yoshimura and Kazunori Nakamoto and Shintaro Ichikawa},
  journal = {Japanese Journal of Radiology},
  year = {2026},
  month = jul,
  doi = {10.1007/s11604-026-02049-8},
  url = {https://doi.org/10.1007/s11604-026-02049-8},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - Open-source offline-deployable retrieval-augmented large language model for assisting pancreatic cancer staging
AU  - Hisashi Johno
AU  - Akitomo Amakawa
AU  - Atsushi Komaba
AU  - Ryota Tozuka
AU  - Yuki Johno
AU  - Junichi Sato
AU  - Kentaro Yoshimura
AU  - Kazunori Nakamoto
AU  - Shintaro Ichikawa
PY  - 2026
DA  - 2026/07/09
JO  - Japanese Journal of Radiology
DO  - 10.1007/s11604-026-02049-8
UR  - https://doi.org/10.1007/s11604-026-02049-8
AB  - Purpose: Large language models (LLMs) are increasingly applied in radiology, but key challenges remain, including data leakage from cloud-based systems, false outputs, and limited reasoning transparency. This study aimed to develop an open-source, offline-deployable retrieval-augmented LLM (RA-LLM) system in which local execution prevents data leakage and retrieval-augmented generation (RAG) improves output accuracy and transparency using reliable external knowledge (REK), demonstrated in pancreatic cancer staging.  Materials and methods: Llama-3.2 11B and Gemma-3 27B were used as local LLMs, and GPT-4o mini served as a cloud-based comparator. The Japanese pancreatic cancer guideline, written in English, served as REK. Relevant REK excerpts were retrieved to generate retrieval-augmented responses. System performance, including classification accuracy, retrieval metrics, and execution time, was evaluated on 100 simulated pancreatic cancer cases with CT findings described in English, using non-RAG LLMs as baselines. McNemar tests were applied to TNM staging and resectability classification.  Results: RAG improved TNM staging accuracy for all LLMs (GPT-4o mini 61% → 90%, p < 0.001; Llama-3.2 11B 53% → 72%, p < 0.001; Gemma-3 27B 59% → 87%, p < 0.001) and mildly improved resectability classification for GPT-4o mini and Llama-3.2 11B (72% → 84%, p = 0.012; 58% → 73%, p = 0.006), while the improvement for Gemma-3 27B was not evident (77% → 86%, p = 0.093). Gemma-3 27B showed performance comparable to GPT-4o mini. Retrieval performance was high (context recall = 1; context precision = 0.5–1), and local models ran at speeds comparable to the cloud-based GPT-4o mini.  Conclusion: We developed an offline-deployable RA-LLM system for pancreatic cancer staging and publicly released its full source code. RA-LLMs outperformed baseline LLMs, and the offline-capable Gemma-3 27B performed comparably to the widely used cloud-based GPT-4o mini.  Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1007/s11604-026-02049-8.
SN  - 1867-1071
SN  - 1867-108X
ER  -