{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,15]],"date-time":"2026-10-15T13:45:29Z","timestamp":1792071929099,"version":"4.4.0"},"reference-count":35,"publisher":"Wiley","issue":"9","license":[{"start":{"date-parts":[[2026,9,10]],"date-time":"2026-09-10T00:00:00Z","timestamp":1788998400000},"content-version":"vor","delay-in-days":9,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0\/"},{"start":{"date-parts":[[2026,9,1]],"date-time":"2026-09-01T00:00:00Z","timestamp":1788220800000},"content-version":"tdm","delay-in-days":0,"URL":"http:\/\/doi.wiley.com\/10.1002\/tdm_license_1.1"}],"content-domain":{"domain":["aapm.onlinelibrary.wiley.com"],"crossmark-restriction":true},"short-container-title":["Medical Physics"],"published-print":{"date-parts":[[2026,9]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Radiotherapy planning is time\u2010intensive, iterative, and operator\u2010dependent. Locally deployed large language\u2010model agents may support planning automation while avoiding transmission of patient data to external model providers.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Purpose<\/jats:title>\n                    <jats:p>The purpose of this work was to develop and evaluate SAGE, a locally deployed LLM agent for autonomous VMAT optimization\u2010priority tuning, and to investigate the impact of model size (8B vs. 70B) and intra\u2010case memory implemented through retrieval augmentation.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>The SAGE agent (Llama 3.1 8B\/70B) was interfaced with a clinical TPS to tune optimization priority values for 17 retrospective prostate cancer patients (60\u00a0Gy\/20fx). All SAGE priority values were initialized to 1, while the study\u2010defined optimization objective set and beam geometry were held fixed. We compared four agent variants (8B, 70B, 8B\u2010RAG, 70B\u2010RAG) to the clinical plan using paired Wilcoxon tests with FDR control.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      No statistically significant difference in target (PTV V58.5\u00a0Gy) coverage was observed between SAGE variants and clinical plans. The 70B\u2010RAG agent significantly improved penile bulb sparing (V22Gy) over clinical plans (\n                      <jats:italic>p<\/jats:italic>\n                      \u00a0=\u00a00.014). Enabling intra\u2010case memory provided a statistically significant improvement in penile bulb sparing, while no significant effect was observed from model size alone. While clinical plans were statistically better for some OARs, all SAGE 70B\u2010RAG plans remained clinically acceptable. The most complex version of SAGE (70B\u2010RAG) completed ten autonomous iterations in a median compute time of 45 min (IQR: 34\u201351). For reference, the clinical planning workflow turnaround (ARIA\u2010recorded task duration) median time was 2700 min (IQR: 2040\u20133060).\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusions<\/jats:title>\n                    <jats:p>This proof\u2010of\u2010concept study demonstrates that a local LLM agent can autonomously tune optimization priority values to generate clinically acceptable prostate VMAT plans within a fixed, study\u2010defined objective space. Intra\u2010case memory produced a statistically significant, endpoint\u2010specific improvement (penile bulb V22Gy). These findings suggest that smaller, efficient models may be viable for constrained treatment\u2010planning optimization tasks.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.1002\/mp.70654","type":"journal-article","created":{"date-parts":[[2026,9,10]],"date-time":"2026-09-10T07:43:08Z","timestamp":1789026188000},"update-policy":"https:\/\/doi.org\/10.1002\/crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["Local large language model\u2010guided optimization for prostate radiotherapy planning"],"prefix":"10.1002","volume":"53","author":[{"given":"Humza","family":"Nusrat","sequence":"first","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"},{"name":"College of Human Medicine Michigan State University  East Lansing Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Bing","family":"Luo","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Hassan","family":"Bagher\u2010Ebadian","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"},{"name":"College of Human Medicine Michigan State University  East Lansing Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ryan","family":"Hall","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Joshua","family":"Kim","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Anthony","family":"Doemer","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Khaled","family":"Adil","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mohamed","family":"Elsheikh","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Aharon","family":"Feldman","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Benjamin","family":"Movsas","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"},{"name":"College of Human Medicine Michigan State University  East Lansing Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Kundan","family":"Thind","sequence":"additional","affiliation":[{"name":"Department of Radiation Oncology Henry Ford Health  Detroit Michigan USA"},{"name":"College of Human Medicine Michigan State University  East Lansing Michigan USA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"311","published-online":{"date-parts":[[2026,9,10]]},"reference":[{"key":"e_1_2_10_2_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.prro.2016.04.005"},{"key":"e_1_2_10_3_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.ejmp.2019.10.032"},{"key":"e_1_2_10_4_1","doi-asserted-by":"publisher","DOI":"10.1002\/ACM2.13337"},{"key":"e_1_2_10_5_1","doi-asserted-by":"publisher","DOI":"10.1002\/MP.13262"},{"key":"e_1_2_10_6_1","doi-asserted-by":"publisher","DOI":"10.1093\/bjro\/tzae017"},{"key":"e_1_2_10_7_1","doi-asserted-by":"publisher","DOI":"10.1038\/s42256\u2010019\u20100048\u2010x"},{"key":"e_1_2_10_8_1","unstructured":"Health Insurance Portability and Accountability Act; 1996. Accessed February 24 2025.https:\/\/www.cdc.gov\/phlp\/php\/resources\/health\u2010insurance\u2010portability\u2010and\u2010accountability\u2010act\u2010of\u20101996\u2010hipaa.html"},{"key":"e_1_2_10_9_1","first-page":"2","article-title":"Agentic LLM workflows for generating patient\u2010friendly medical reports","author":"Sudarshan M","year":"2024","journal-title":"arXiv"},{"key":"e_1_2_10_10_1","doi-asserted-by":"publisher","DOI":"10.1038\/s42256\u2010024\u201000944\u20101"},{"key":"e_1_2_10_11_1","doi-asserted-by":"crossref","unstructured":"ZhaoX LiuS YangSY et\u00a0al.Enhancing retrieval\u2010augmented generation with knowledge graph\u2010elicited reasoning for healthcare copilot.WWW 2025 \u2010 Proceedings of the ACM Web Conference.2025;1:4442\u20104457. doi:10.1145\/3696410.3714782","DOI":"10.1145\/3696410.3714782"},{"key":"e_1_2_10_12_1","doi-asserted-by":"publisher","DOI":"10.1101\/2025.09.16.25335813"},{"key":"e_1_2_10_13_1","doi-asserted-by":"publisher","DOI":"10.1101\/2023.02.21.23286254"},{"key":"e_1_2_10_14_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.adro.2023.101400"},{"key":"e_1_2_10_15_1","first-page":"1","article-title":"RadOnc\u2010GPT: a large language model for radiation oncology","author":"Liu Z","year":"2023","journal-title":"arXiv"},{"key":"e_1_2_10_16_1","doi-asserted-by":"publisher","DOI":"10.1038\/s41467\u2010024\u201053387\u2010y"},{"key":"e_1_2_10_17_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.ijrobp.2025.06.3149"},{"key":"e_1_2_10_18_1","first-page":"1","article-title":"Autonomous radiotherapy treatment planning using DOLA: a privacy\u2010preserving, LLM\u2010based optimization agent","author":"Nusrat H","year":"2025","journal-title":"arXiv"},{"key":"e_1_2_10_19_1","first-page":"1","article-title":"Zero\u2010shot large language model agents for fully automated radiotherapy treatment planning","author":"Yang D","year":"2025","journal-title":"arXiv"},{"key":"e_1_2_10_20_1","doi-asserted-by":"publisher","DOI":"10.1186\/S13014\u2010025\u201002660\u20105\/TABLES\/5"},{"key":"e_1_2_10_21_1","doi-asserted-by":"publisher","DOI":"10.1088\/1361\u20106560\/ADF02C"},{"key":"e_1_2_10_22_1","doi-asserted-by":"publisher","DOI":"10.1088\/1361\u20106560\/adbff1"},{"key":"e_1_2_10_23_1","doi-asserted-by":"publisher","DOI":"10.21203\/RS.3.RS\u20104240043\/V1"},{"key":"e_1_2_10_24_1","doi-asserted-by":"publisher","DOI":"10.52202\/075280-2400"},{"key":"e_1_2_10_25_1","first-page":"1","article-title":"MEDITRON\u201070B: scaling medical pretraining for large language models","author":"Chen Z","year":"2023","journal-title":"arXiv"},{"key":"e_1_2_10_26_1","doi-asserted-by":"publisher","DOI":"10.1038\/s41746\u2010025\u201001519\u2010z"},{"key":"e_1_2_10_27_1","doi-asserted-by":"crossref","unstructured":"XiongG JinQ LuZ ZhangA Benchmarking retrieval\u2010augmented generation for medicine.Proceedings of the Annual Meeting of the Association for Computational Linguistics. Published online February 20 2024:6233\u20106251. doi:10.18653\/v1\/2024.findings\u2010acl.372","DOI":"10.18653\/v1\/2024.findings-acl.372"},{"key":"e_1_2_10_28_1","doi-asserted-by":"publisher","DOI":"10.1200\/JCO.2016.71.7397"},{"key":"e_1_2_10_29_1","first-page":"1","article-title":"The Llama 3 herd of models","author":"Grattafiori A","year":"2024","journal-title":"arXiv"},{"key":"e_1_2_10_30_1","first-page":"1","article-title":"LLaMA: open and efficient foundation language models","author":"Touvron H","year":"2023","journal-title":"arXiv"},{"key":"e_1_2_10_31_1","first-page":"8","article-title":"Scaling language models: methods, analysis & insights from training gopher","author":"Rae JW","year":"2021","journal-title":"arXiv"},{"key":"e_1_2_10_32_1","first-page":"240","article-title":"PaLM: scaling language modeling with pathways","volume":"24","author":"Chowdhery A","year":"2022","journal-title":"J Mach Learn Res"},{"key":"e_1_2_10_33_1","doi-asserted-by":"publisher","DOI":"10.52202\/068431-2176"},{"key":"e_1_2_10_34_1","doi-asserted-by":"crossref","unstructured":"RobertsA RaffelC ShazeerN.How much knowledge can you pack into the parameters of a language model?.EMNLP 2020 \u2010 2020 Conference on Empirical Methods in Natural Language Processing.2020:5418\u20105426. doi:10.18653\/v1\/2020.emnlp\u2010main.437","DOI":"10.18653\/v1\/2020.emnlp-main.437"},{"key":"e_1_2_10_35_1","first-page":"14","article-title":"Scaling laws for neural language models","author":"Kaplan J","year":"2020","journal-title":"arXiv"},{"key":"e_1_2_10_36_1","first-page":"1877","article-title":"Language models are few\u2010shot learners","volume":"33","author":"Brown TB","year":"2020","journal-title":"Adv Neural Inf Process Syst"}],"container-title":["Medical Physics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/pdf\/10.1002\/mp.70654","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/full-xml\/10.1002\/mp.70654","content-type":"application\/xml","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/pdf\/10.1002\/mp.70654","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,15]],"date-time":"2026-10-15T12:55:34Z","timestamp":1792068934000},"score":1,"resource":{"primary":{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/10.1002\/mp.70654"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9]]},"references-count":35,"journal-issue":{"issue":"9","published-print":{"date-parts":[[2026,9]]}},"alternative-id":["10.1002\/mp.70654"],"URL":"https:\/\/doi.org\/10.1002\/mp.70654","archive":["Portico"],"relation":{},"ISSN":["0094-2405","2473-4209"],"issn-type":[{"value":"0094-2405","type":"print"},{"value":"2473-4209","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9]]},"assertion":[{"value":"2025-11-25","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-07-26","order":2,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-09-10","order":3,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}],"article-number":"e70654"}}