Automated stereotactic radiosurgery planning using a human-in-the-loop reasoning large language model agent

Automated stereotactic radiosurgery planning using a human-in-the-loop reasoning large language model agent | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Automated stereotactic radiosurgery planning using a human-in-the-loop reasoning large language model agent Humza Nusrat, Luke Francisco, Bing Luo, Hassan Bagher-Ebadian, and 9 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8612138/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Stereotactic radiosurgery (SRS) demands precise dose shaping around critical structures, yet black-box AI systems have limited clinical adoption due to opacity concerns. We tested whether chain-of-thought reasoning improves agentic planning in a retrospective cohort of 41 patients with brain metastases treated with 18 Gy single-fraction SRS. We developed SAGE (Secure Agent for Generative Dose Expertise), an LLM-based planning agent for automated SRS treatment planning. Two variants generated plans for each case: one using a non-reasoning model, one using a reasoning model. The reasoning variant showed comparable plan dosimetry relative to human planners on primary endpoints (PTV coverage, maximum dose, conformity index, gradient index; all $p$ \textgreater\0.21) while reducing cochlear dose below human baselines ($p$ = 0.022). When prompted to improve conformity, the reasoning model demonstrated systematic planning behaviors including prospective constraint verification (457 instances) and trade-off deliberation (609 instances), while the standard model exhibited none of these deliberative processes (0 and 7 instances, respectively). Content analysis revealed that constraint verification and causal explanation concentrated in the reasoning agent. The optimization traces serve as auditable logs, offering a path toward transparent automated planning. Biological sciences/Cancer Health sciences/Medical research Health sciences/Oncology Reasoning model Radiotherapy planning Agentic AI Full Text Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterials.pdf Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8612138","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":578266400,"identity":"f72e65f7-eb85-423f-b987-0bf9689b48e8","order_by":0,"name":"Humza Nusrat","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxUlEQVRIiWNgGAWjYBACAyjNwy/B+IwkLQYykjOYzUjTYmNwg1gt5tI9xp8rav7wGN9uZnv4g8EmX96BgBbLOWfMJM8cM+Axu3OY3ZiHIc1y4wFCDruRY8bYwAbUciP/mDQDw2EDwwbCWow/Nvwz4DGekcwm+YNILQaSjW0GPAYSyWwSPEAt8gR0ALWklUk29hnzSNxIZpPmMUgzMCCsJXnzx4Zvcvb8YIdV2BjIE3IYuglAdIA0LUBAqi2jYBSMglEw/AEA2Q44iITKbNsAAAAASUVORK5CYII=","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":true,"prefix":"","firstName":"Humza","middleName":"","lastName":"Nusrat","suffix":""},{"id":578266401,"identity":"b66ecda6-d831-4ae7-bb1f-871a23d5ebcb","order_by":1,"name":"Luke Francisco","email":"","orcid":"","institution":"University of Michigan–Ann Arbor","correspondingAuthor":false,"prefix":"","firstName":"Luke","middleName":"","lastName":"Francisco","suffix":""},{"id":578266402,"identity":"ba12610b-bda7-4769-bfc0-31408d247afd","order_by":2,"name":"Bing Luo","email":"","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":false,"prefix":"","firstName":"Bing","middleName":"","lastName":"Luo","suffix":""},{"id":578266403,"identity":"e1295d3e-1715-4e13-bb5e-322365c82dc1","order_by":3,"name":"Hassan Bagher-Ebadian","email":"","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":false,"prefix":"","firstName":"Hassan","middleName":"","lastName":"Bagher-Ebadian","suffix":""},{"id":578266404,"identity":"4815207c-06ac-4a6e-ae83-dff411d37c3f","order_by":4,"name":"Joshua Kim","email":"","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":false,"prefix":"","firstName":"Joshua","middleName":"","lastName":"Kim","suffix":""},{"id":578266405,"identity":"9e413b18-3c57-4ae8-8b62-8dd9eed9f7b0","order_by":5,"name":"Karen Chin Snyder","email":"","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":false,"prefix":"","firstName":"Karen","middleName":"Chin","lastName":"Snyder","suffix":""},{"id":578266406,"identity":"810a2a4d-b0bc-4be3-ae9d-67dace091224","order_by":6,"name":"Salim Siddiqui","email":"","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":false,"prefix":"","firstName":"Salim","middleName":"","lastName":"Siddiqui","suffix":""},{"id":578266407,"identity":"1a16e443-58b5-4fda-bd10-0ec6899c62e9","order_by":7,"name":"Mira Shah","email":"","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":false,"prefix":"","firstName":"Mira","middleName":"","lastName":"Shah","suffix":""},{"id":578266408,"identity":"4f8d640a-0112-401f-ad30-8f074f3c91fb","order_by":8,"name":"Eric Mellon","email":"","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":false,"prefix":"","firstName":"Eric","middleName":"","lastName":"Mellon","suffix":""},{"id":578266409,"identity":"203aee8d-aa97-44b4-8c95-e7e5e87dc35d","order_by":9,"name":"Mohammad Ghassemi","email":"","orcid":"","institution":"Michigan State University","correspondingAuthor":false,"prefix":"","firstName":"Mohammad","middleName":"","lastName":"Ghassemi","suffix":""},{"id":578266410,"identity":"9d17b2f1-1b91-4bc4-98f7-6f7d92acafb7","order_by":10,"name":"Anthony Doemer","email":"","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":false,"prefix":"","firstName":"Anthony","middleName":"","lastName":"Doemer","suffix":""},{"id":578266411,"identity":"a2d79dee-a092-40ea-baae-67c479fc51d7","order_by":11,"name":"Benjamin Movsas","email":"","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":false,"prefix":"","firstName":"Benjamin","middleName":"","lastName":"Movsas","suffix":""},{"id":578266412,"identity":"000fc468-ce1f-402e-a032-357c2db25137","order_by":12,"name":"Kundan Thind","email":"","orcid":"","institution":"Henry Ford Health System","correspondingAuthor":false,"prefix":"","firstName":"Kundan","middleName":"","lastName":"Thind","suffix":""}],"badges":[],"createdAt":"2026-01-15 15:54:04","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8612138/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8612138/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":100939557,"identity":"870439d5-eea8-41dc-9c3b-1dacc31aa40c","added_by":"auto","created_at":"2026-01-23 04:09:52","extension":"json","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13193,"visible":true,"origin":"","legend":"","description":"","filename":"0283b7726f1641ebae7b09cf4435aeeb.json","url":"https://assets-eu.researchsquare.com/files/rs-8612138/v1/36ef967b0a81dd95f6aa4c95.json"},{"id":108509779,"identity":"b2d75015-a388-4996-9647-bda7171e0d0e","added_by":"auto","created_at":"2026-05-05 12:27:47","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4876289,"visible":true,"origin":"","legend":"","description":"","filename":"latex7.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8612138/v1_covered_9c34c799-4116-46cb-9b34-30e73e31256c.pdf"},{"id":100939553,"identity":"993b0ee8-7464-4eb2-b75b-755dc51cbf08","added_by":"auto","created_at":"2026-01-23 04:09:50","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":227748,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterials.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8612138/v1/cdb85583419b6241c486a662.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Automated stereotactic radiosurgery planning using a human-in-the-loop reasoning large language model agent","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Reasoning model, Radiotherapy planning, Agentic AI","lastPublishedDoi":"10.21203/rs.3.rs-8612138/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8612138/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Stereotactic radiosurgery (SRS) demands precise dose shaping around critical structures, yet black-box AI systems have limited clinical adoption due to opacity concerns. We tested whether chain-of-thought reasoning improves agentic planning in a retrospective cohort of 41 patients with brain metastases treated with 18 Gy single-fraction SRS. We developed SAGE (Secure Agent for Generative Dose Expertise), an LLM-based planning agent for automated SRS treatment planning. Two variants generated plans for each case: one using a non-reasoning model, one using a reasoning model. The reasoning variant showed comparable plan dosimetry relative to human planners on primary endpoints (PTV coverage, maximum dose, conformity index, gradient index; all $p$ \\textgreater\\0.21) while reducing cochlear dose below human baselines ($p$ = 0.022). When prompted to improve conformity, the reasoning model demonstrated systematic planning behaviors including prospective constraint verification (457 instances) and trade-off deliberation (609 instances), while the standard model exhibited none of these deliberative processes (0 and 7 instances, respectively). Content analysis revealed that constraint verification and causal explanation concentrated in the reasoning agent. The optimization traces serve as auditable logs, offering a path toward transparent automated planning.","manuscriptTitle":"Automated stereotactic radiosurgery planning using a human-in-the-loop reasoning large language model agent","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-23 04:09:25","doi":"10.21203/rs.3.rs-8612138/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"b8e54d8b-4115-4fa6-b66c-fdfb65c756b0","owner":[],"postedDate":"January 23rd, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Rejected","date":"2026-05-05T12:10:51+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":61528018,"name":"Biological sciences/Cancer"},{"id":61528019,"name":"Health sciences/Medical research"},{"id":61528020,"name":"Health sciences/Oncology"}],"tags":[],"updatedAt":"2026-05-05T12:26:14+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-23 04:09:25","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8612138","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8612138","identity":"rs-8612138","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}