{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,4,28]],"date-time":"2026-04-28T07:23:01Z","timestamp":1777360981210,"version":"3.51.4"},"reference-count":25,"publisher":"Springer Science and Business Media LLC","issue":"1","license":[{"start":{"date-parts":[[2025,7,2]],"date-time":"2025-07-02T00:00:00Z","timestamp":1751414400000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2025,7,2]],"date-time":"2025-07-02T00:00:00Z","timestamp":1751414400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"funder":[{"DOI":"10.13039\/501100001691","name":"Japan Society for the Promotion of Science","doi-asserted-by":"publisher","award":["25K02609"],"award-info":[{"award-number":["25K02609"]}],"id":[{"id":"10.13039\/501100001691","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001691","name":"Japan Society for the Promotion of Science","doi-asserted-by":"publisher","award":["25K02609"],"award-info":[{"award-number":["25K02609"]}],"id":[{"id":"10.13039\/501100001691","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001691","name":"Japan Society for the Promotion of Science","doi-asserted-by":"publisher","award":["25K02609"],"award-info":[{"award-number":["25K02609"]}],"id":[{"id":"10.13039\/501100001691","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001691","name":"Japan Society for the Promotion of Science","doi-asserted-by":"publisher","award":["25K02609"],"award-info":[{"award-number":["25K02609"]}],"id":[{"id":"10.13039\/501100001691","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["npj Digit. Med."],"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Large language models (LLMs) demonstrate significant potential in healthcare applications, but clinical deployment is limited by privacy concerns and insufficient medical domain training. This study investigated whether retrieval-augmented generation (RAG) can improve locally deployable LLM for radiology contrast media consultation. In 100 synthetic iodinated contrast media consultations we compared Llama 3.2-11B (baseline and RAG) with three cloud-based models\u2014GPT-4o mini, Gemini 2.0 Flash and Claude 3.5 Haiku. A blinded radiologist ranked the five replies per case, and three LLM-based judges scored accuracy, safety, structure, tone, applicability and latency. Under controlled conditions, RAG eliminated hallucinations (0% vs 8%; \u03c7\u00b2\u208dYates\u208e = 6.38, p\u2009=\u20090.012) and improved mean rank by 1.3 (Z = \u20134.82, p\u2009&lt;\u20090.001), though performance gaps with cloud models persist. The RAG-enhanced model remained faster (2.6\u2009s vs 4.9\u20137.3\u2009s) while the LLM-based judges preferred it over GPT-4o mini, though the radiologist ranked GPT-4o mini higher. RAG thus provides meaningful improvements for local clinical LLMs while maintaining the privacy benefits of on-premise deployment.<\/jats:p>","DOI":"10.1038\/s41746-025-01802-z","type":"journal-article","created":{"date-parts":[[2025,7,2]],"date-time":"2025-07-02T01:40:40Z","timestamp":1751420440000},"update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":20,"title":["Retrieval-augmented generation elevates local LLM quality in radiology contrast media consultation"],"prefix":"10.1038","volume":"8","author":[{"given":"Akihiko","family":"Wada","sequence":"first","affiliation":[]},{"given":"Yuya","family":"Tanaka","sequence":"additional","affiliation":[]},{"given":"Mitsuo","family":"Nishizawa","sequence":"additional","affiliation":[]},{"given":"Akira","family":"Yamamoto","sequence":"additional","affiliation":[]},{"given":"Toshiaki","family":"Akashi","sequence":"additional","affiliation":[]},{"given":"Akifumi","family":"Hagiwara","sequence":"additional","affiliation":[]},{"given":"Yayoi","family":"Hayakawa","sequence":"additional","affiliation":[]},{"given":"Junko","family":"Kikuta","sequence":"additional","affiliation":[]},{"given":"Keigo","family":"Shimoji","sequence":"additional","affiliation":[]},{"given":"Katsuhiro","family":"Sano","sequence":"additional","affiliation":[]},{"given":"Koji","family":"Kamagata","sequence":"additional","affiliation":[]},{"given":"Atsushi","family":"Nakanishi","sequence":"additional","affiliation":[]},{"given":"Shigeki","family":"Aoki","sequence":"additional","affiliation":[]}],"member":"297","published-online":{"date-parts":[[2025,7,2]]},"reference":[{"key":"1802_CR1","unstructured":"Liu, Z. et al. Radiology-GPT: A Large Language Model for Radiology. Preprint at arXiv:2306.08666 (2023)."},{"key":"1802_CR2","first-page":"1","volume":"37","author":"T Nakaura","year":"2024","unstructured":"Nakaura, T. et al. The impact of large language models on radiology: a guide for radiologists on the latest innovations in AI. Jpn. J. Radiol. 37, 1\u201312 (2024).","journal-title":"Jpn. J. Radiol."},{"key":"1802_CR3","doi-asserted-by":"publisher","first-page":"1541","DOI":"10.3390\/diagnostics14141541","volume":"14","author":"A Wada","year":"2024","unstructured":"Wada, A. et al. Optimizing GPT-4 Turbo Diagnostic Accuracy in Neuroradiology through Prompt Engineering and Confidence Thresholds. Diagnostics 14, 1541 (2024).","journal-title":"Diagnostics"},{"key":"1802_CR4","doi-asserted-by":"publisher","first-page":"278","DOI":"10.60087\/jklst.vol3.n3.p.278-287","volume":"3","author":"N Abbasi","year":"2024","unstructured":"Abbasi, N. & Smith, D. A. Cybersecurity in Healthcare: Securing Patient Health Information (PHI). J. Knowl. Learn. Sci. Technol. 3, 278\u2013287 (2024).","journal-title":"J. Knowl. Learn. Sci. Technol."},{"key":"1802_CR5","doi-asserted-by":"crossref","unstructured":"Mohan, A. et al. Securing AI Inference in the Cloud: Is CPU-GPU Confidential Computing Ready? In Proc. 17th IEEE Int. Conf. Cloud Comput. 164-175 (IEEE, 2024).","DOI":"10.1109\/CLOUD62652.2024.00028"},{"key":"1802_CR6","doi-asserted-by":"crossref","unstructured":"Bedi, S. et al. A Systematic Review of Testing and Evaluation of Healthcare Applications of Large Language Models (LLMs). Preprint at medRxiv:2024.04.15.24305869 (2024).","DOI":"10.1101\/2024.04.15.24305869"},{"key":"1802_CR7","first-page":"9459","volume":"33","author":"P Lewis","year":"2020","unstructured":"Lewis, P. et al. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. Advances in Neural Information Processing Systems 33, 9459\u20139474 (2020).","journal-title":"Advances in Neural Information Processing Systems"},{"key":"1802_CR8","doi-asserted-by":"crossref","unstructured":"Nguyen, Q. et al. Advancing Question-Answering in Ophthalmology with Retrieval-Augmented Generation (RAG): Benchmarking Open-source and Proprietary Large Language Models. Preprint at medRxiv:2024.11.18.24317510 (2024).","DOI":"10.1101\/2024.11.18.24317510"},{"key":"1802_CR9","unstructured":"Mortaheb, M., Khojastepour, M. A. A., Chakradhar, S. T. & Ulukus, S. RAG-Check: Evaluating Multimodal Retrieval Augmented Generation Performance. Preprint at arXiv:2501.03995 (2025)."},{"key":"1802_CR10","doi-asserted-by":"publisher","DOI":"10.1111\/eci.14183","volume":"54","author":"A Telenti","year":"2024","unstructured":"Telenti, A. et al. Large language models for science and medicine. Eur. J. Clin. Investig. 54, e14183 (2024).","journal-title":"Eur. J. Clin. Investig."},{"key":"1802_CR11","doi-asserted-by":"publisher","first-page":"476","DOI":"10.1001\/jama.2022.9879","volume":"328","author":"MS Davenport","year":"2022","unstructured":"Davenport, M. S., Chu, P., Szczykutowicz, T. P. & Smith-Bindman, R. Comparison of Strategies to Conserve Iodinated Intravascular Contrast Media for Computed Tomography During a Shortage. JAMA 328, 476\u2013478 (2022).","journal-title":"JAMA"},{"key":"1802_CR12","doi-asserted-by":"crossref","unstructured":"CADTH. Optimizing the Use of Iodinated Contrast Media for CT: Managing Shortages and Planning for a Sustainable and Secure Supply. CADTH report 613 (2023).","DOI":"10.51731\/cjht.2023.613"},{"key":"1802_CR13","doi-asserted-by":"publisher","first-page":"51","DOI":"10.1080\/14740338.2016.1226280","volume":"15","author":"M Sessa","year":"2016","unstructured":"Sessa, M. et al. Campania preventability assessment committee: a focus on the preventability of the contrast media adverse drug reactions. Expert Opin. Drug Saf. 15, 51\u201359 (2016).","journal-title":"Expert Opin. Drug Saf."},{"key":"1802_CR14","doi-asserted-by":"publisher","first-page":"573","DOI":"10.2214\/AJR.15.15117","volume":"206","author":"B Siewert","year":"2016","unstructured":"Siewert, B., Brook, O. R., Hochman, M. & Eisenberg, R. L. Impact of Communication Errors in Radiology on Patient Care, Customer Satisfaction, and Work-Flow Efficiency. Am. J. Roentgenol. 206, 573\u2013579 (2016).","journal-title":"Am. J. Roentgenol."},{"key":"1802_CR15","unstructured":"Gupta, S., Ranjan, R. & Singh, S. N. A Comprehensive Survey of Retrieval-Augmented Generation. Preprint at arXiv:2410.12837 (2024)."},{"key":"1802_CR16","doi-asserted-by":"publisher","DOI":"10.2196\/53164","volume":"26","author":"M Chelli","year":"2024","unstructured":"Chelli, M. et al. Hallucination Rates and Reference Accuracy of ChatGPT and Bard for Systematic Reviews: Comparative Analysis. J. Med. Internet Res. 26, e53164 (2024).","journal-title":"J. Med. Internet Res."},{"key":"1802_CR17","doi-asserted-by":"publisher","first-page":"586","DOI":"10.1093\/jamia\/ocaf005","volume":"32","author":"MY Ng","year":"2025","unstructured":"Ng, M. Y., Helzer, J., Pfeffer, M. A., Seto, T. & Hernandez-Boussard, T. Development of secure infrastructure for advancing generative artificial intelligence research in healthcare at an academic medical center. J. Am. Med. Inform. Assoc. 32, 586\u2013588 (2025).","journal-title":"J. Am. Med. Inform. Assoc."},{"key":"1802_CR18","unstructured":"Iglesia, I. D. et al. Ranking Over Scoring: Towards Reliable and Robust Automated Evaluation of LLM-Generated Medical Explanatory Arguments. In Proc. 31st Int. Conf. Computational Linguistics (COLING), 2025.coling-main.634 (2025)."},{"key":"1802_CR19","unstructured":"Gu, J. et al. A Survey on LLM-as-a-Judge. Preprint at arXiv:2411.15594 (2024)."},{"key":"1802_CR20","doi-asserted-by":"publisher","unstructured":"Szymanski, A. et al. Limitations of the LLM-as-a-Judge Approach for Evaluating LLM Outputs in Expert Knowledge Tasks. In Proc. 30th Int. Conf. Intelligent User Interfaces (IUI), 952\u2212966. (ACM, 2025). https:\/\/doi.org\/10.1145\/3708359.3712091.","DOI":"10.1145\/3708359.3712091"},{"key":"1802_CR21","unstructured":"Dify.AI. The Innovation Engine for Generative AI Applications. https:\/\/dify.ai\/ (2023)."},{"key":"1802_CR22","doi-asserted-by":"publisher","first-page":"1738","DOI":"10.1148\/rg.2015150033","volume":"35","author":"KR Beckett","year":"2015","unstructured":"Beckett, K. R., Moriarity, A. K. & Langer, J. M. Safe Use of Contrast Media: What the Radiologist Needs to Know. Radiographics 35, 1738\u20131750 (2015).","journal-title":"Radiographics"},{"key":"1802_CR23","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1007\/s10157-019-01750-5","volume":"24","author":"Y Isaka","year":"2020","unstructured":"Isaka, Y. et al. Guideline on the use of iodinated contrast media in patients with kidney disease 2018. Clin. Exp. Nephrol. 24, 1\u201344 (2020).","journal-title":"Clin. Exp. Nephrol."},{"key":"1802_CR24","doi-asserted-by":"publisher","first-page":"131","DOI":"10.1007\/s00261-005-0380-y","volume":"31","author":"HS Thomsen","year":"2006","unstructured":"Thomsen, H. S. & Morcos, S. K. & ESUR. ESUR guidelines on contrast media. Abdom. Imaging 31, 131\u2013140 (2006).","journal-title":"Abdom. Imaging"},{"key":"1802_CR25","doi-asserted-by":"crossref","unstructured":"Liu, Y. et al. G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment. In Proc. Conf. Empirical Methods in Natural Language Processing (EMNLP), 2511\u20132522 (2023).","DOI":"10.18653\/v1\/2023.emnlp-main.153"}],"container-title":["npj Digital Medicine"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.nature.com\/articles\/s41746-025-01802-z.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s41746-025-01802-z","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s41746-025-01802-z.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,7,2]],"date-time":"2025-07-02T01:40:42Z","timestamp":1751420442000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.nature.com\/articles\/s41746-025-01802-z"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,7,2]]},"references-count":25,"journal-issue":{"issue":"1","published-online":{"date-parts":[[2025,12]]}},"alternative-id":["1802"],"URL":"https:\/\/doi.org\/10.1038\/s41746-025-01802-z","relation":{"has-preprint":[{"id-type":"doi","id":"10.21203\/rs.3.rs-6538325\/v1","asserted-by":"object"}]},"ISSN":["2398-6352"],"issn-type":[{"value":"2398-6352","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,7,2]]},"assertion":[{"value":"27 April 2025","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"15 June 2025","order":2,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"2 July 2025","order":3,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"The authors declare no competing interests.","order":1,"name":"Ethics","group":{"name":"EthicsHeading","label":"Competing interests"}}],"article-number":"395"}}