{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,21]],"date-time":"2026-07-21T04:49:01Z","timestamp":1784609341344,"version":"3.55.0"},"reference-count":14,"publisher":"Wiley","issue":"2","license":[{"start":{"date-parts":[[2026,1,29]],"date-time":"2026-01-29T00:00:00Z","timestamp":1769644800000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/onlinelibrary.wiley.com\/termsAndConditions#vor"},{"start":{"date-parts":[[2026,1,29]],"date-time":"2026-01-29T00:00:00Z","timestamp":1769644800000},"content-version":"tdm","delay-in-days":0,"URL":"http:\/\/doi.wiley.com\/10.1002\/tdm_license_1.1"}],"funder":[{"DOI":"10.13039\/100031478","name":"NextGenerationEU","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100031478","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100010661","name":"Horizon 2020 Framework Programme","doi-asserted-by":"publisher","award":["101017727"],"award-info":[{"award-number":["101017727"]}],"id":[{"id":"10.13039\/100010661","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["aapm.onlinelibrary.wiley.com"],"crossmark-restriction":true},"short-container-title":["Medical Physics"],"published-print":{"date-parts":[[2026,2]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Radiotherapy treatment planning (TP) aims to maximize radiation dose delivered to tumors while minimizing exposure to surrounding healthy tissues. Beam angle optimization (BAO) is a crucial component of TP, characterized by high dimensionality and non\u2010convexity, and is traditionally solved via heuristic or manual iterative approaches. These conventional methods are time\u2010consuming and often yield suboptimal solutions due to incomplete exploration of the vast solution\u00a0space.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Purpose<\/jats:title>\n                    <jats:p>This study introduces a novel framework integrating a general\u2010purpose large language model (LLM) within a reinforcementlearning (RL)\u2010inspired iterative strategy to automate BAO in radiotherapy planning. Taking advantage of the inherent knowledge embedded in LLMs, the method uses visual and scalar feedback to produce clinically meaningful treatment plans without requiring any domain\u2010specific fine\u2010tuning or additional\u00a0training.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>\n                      The proposed framework employs an off\u2010the\u2010shelf Generative Pre\u2010trained Transformer, GPT\u20104 model (denoted GPT\u20104\n                      <jats:sub>o<\/jats:sub>\n                      ) in an inference\u2010only setting. At each iteration, GPT\u20104\n                      <jats:sub>o<\/jats:sub>\n                      suggests a set of gantry angles, which are subsequently input into the\n                      <jats:sc>MatRad<\/jats:sc>\n                      software to generate a dose distribution. A scalar reward is computed from this distribution using a custom reward function designed to balance target dose conformity and sparing of organs\u2010at\u2010risk (OARs). This reward, along with the corresponding dose maps, serves as feedback for the LLM to iteratively refine its suggestions. The refinement process consists of distinct exploration and exploitation phases inspired by classical RL paradigms. We evaluated six configurations that varied in exploration duration and in the Computed Tomography (CT) slice inputs provided to the LLM (Single\u2010View vs. Multi\u2010View). Performance was benchmarked against a random\u2010angle selection baseline across three anatomical sites: prostate, head\u2010and\u2010neck, and\u00a0liver.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>Across the liver and head\u2010and\u2010neck cases, all LLM\u2010based configurations significantly outperformed the random baseline (). In the prostate scenario, most strategies demonstrated statistically significant improvements, except for the Multi\u2010View configurations with extended exploration phases (10 and 15 iterations). Rewards consistently increased during the exploitation phase, and the resulting dose\u2013volume histograms and dose distributions exhibited improved conformity to target volumes with enhanced sparing of OARs. Notably, plans of clinically plausible quality were obtained within 20 iterative refinement steps in this proof\u2010of\u2010concept\u00a0setting.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusions<\/jats:title>\n                    <jats:p>This study demonstrates that general\u2010purpose LLMs, operating without specialized model training or fine\u2010tuning, can effectively serve as intelligent agents for automated radiotherapy TP, specifically addressing the BAO problem. This flexible and scalable framework has the potential to enhance clinical decision\u2010making workflows in radiotherapy. Future research directions include exploring more comprehensive and clinically nuanced reward functions and extending the methodology to other components of radiotherapy\u00a0TP.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.1002\/mp.70258","type":"journal-article","created":{"date-parts":[[2026,1,30]],"date-time":"2026-01-30T04:30:15Z","timestamp":1769747415000},"update-policy":"https:\/\/doi.org\/10.1002\/crossmark_policy","source":"Crossref","is-referenced-by-count":1,"title":["Beam angle optimization for radiotherapy using LLMs via reinforcement\u2010learning inspired iterative refinement"],"prefix":"10.1002","volume":"53","author":[{"given":"Sara","family":"Cammarota","sequence":"first","affiliation":[{"name":"Department of Biomedicine and Prevention University of Rome Tor Vergata Rome Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Matteo","family":"Ferrante","sequence":"additional","affiliation":[{"name":"Department of Biomedicine and Prevention University of Rome Tor Vergata Rome Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Alessandra","family":"Carosi","sequence":"additional","affiliation":[{"name":"Radiotherapy Unit Department of Oncology and Hematology Rome Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Rolando Maria","family":"D'Angelillo","sequence":"additional","affiliation":[{"name":"Radiation Oncology, Department of Biomedicine and Prevention University of Rome Tor Vergata Rome Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Nicola","family":"Toschi","sequence":"additional","affiliation":[{"name":"Department of Biomedicine and Prevention University of Rome Tor Vergata Rome Italy"},{"name":"A.A. Martinos Center for Biomedical Imaging Harvard Medical School\/MGH Boston US"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"311","published-online":{"date-parts":[[2026,1,29]]},"reference":[{"key":"e_1_2_9_2_1","doi-asserted-by":"publisher","DOI":"10.7150\/ijms.3635"},{"key":"e_1_2_9_3_1","doi-asserted-by":"publisher","DOI":"10.3389\/fonc.2020.580919"},{"key":"e_1_2_9_4_1","unstructured":"BaoP. WangG. YangR. &DongB.(2023).Deep Reinforcement Learning for Beam Angle Optimization of Intensity\u2010Modulated Radiation Therapy."},{"key":"e_1_2_9_5_1","doi-asserted-by":"publisher","DOI":"10.1088\/0031-9155\/48\/19\/007"},{"key":"e_1_2_9_6_1","doi-asserted-by":"publisher","DOI":"10.1088\/0031-9155\/49\/10\/007"},{"key":"e_1_2_9_7_1","doi-asserted-by":"publisher","DOI":"10.1118\/1.1601911"},{"key":"e_1_2_9_8_1","volume-title":"Reinforcement Learning: An Introduction","author":"Sutton RS","year":"2018"},{"key":"e_1_2_9_9_1","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-031-82007-6_5"},{"key":"e_1_2_9_10_1","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-19387-8_391"},{"key":"e_1_2_9_11_1","first-page":"1877","volume-title":"Language models are few\u2010shot learners. In Proceedings of the 34th International Conference on Neural Information Processing Systems (NIPS '20).","author":"Brown T"},{"key":"e_1_2_9_12_1","doi-asserted-by":"crossref","unstructured":"LiuNF LinK HewittJ et\u00a0al.Lost in the Middle: How Language Models Use Long Contexts.Transactions of the Association for Computational Linguistics 2024.12:157\u2013173.","DOI":"10.1162\/tacl_a_00638"},{"key":"e_1_2_9_13_1","doi-asserted-by":"publisher","DOI":"10.1088\/1361-6560\/adf02c"},{"key":"e_1_2_9_14_1","doi-asserted-by":"publisher","DOI":"10.1088\/1361-6560\/adbff1"},{"key":"e_1_2_9_15_1","unstructured":"NusratH LuoB HallR et\u00a0al.Autonomous radiotherapy treatment planning using DOLA: a privacy\u2010preserving LLM\u2010based optimization agent.2025."}],"container-title":["Medical Physics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/pdf\/10.1002\/mp.70258","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/full-xml\/10.1002\/mp.70258","content-type":"application\/xml","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/pdf\/10.1002\/mp.70258","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,9]],"date-time":"2026-03-09T13:35:14Z","timestamp":1773063314000},"score":1,"resource":{"primary":{"URL":"https:\/\/aapm.onlinelibrary.wiley.com\/doi\/10.1002\/mp.70258"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,1,29]]},"references-count":14,"journal-issue":{"issue":"2","published-print":{"date-parts":[[2026,2]]}},"alternative-id":["10.1002\/mp.70258"],"URL":"https:\/\/doi.org\/10.1002\/mp.70258","archive":["Portico"],"relation":{},"ISSN":["0094-2405","2473-4209"],"issn-type":[{"value":"0094-2405","type":"print"},{"value":"2473-4209","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,1,29]]},"assertion":[{"value":"2025-05-11","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2025-11-23","order":2,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-01-29","order":3,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}],"article-number":"e70258"}}