{"paper":{"title":"Learning to Optimize Radiotherapy Plans via Fluence Maps Diffusion Model Generation and LSTM-based Optimization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A distilled diffusion model generates clinically feasible fluence maps in one shot for VMAT radiotherapy, then an LSTM refines them to meet dose goals.","cross_cats":["eess.IV"],"primary_cat":"cs.CV","authors_text":"Ali Kamen, Dorin Comaniciu, Florin C. Ghesu, Isabella Poles, Marco D. Santambrogio, Martin Kraus, Riqiang Gao, Simon Arberet","submitted_at":"2026-05-13T16:00:11Z","abstract_excerpt":"Volumetric Modulated Arc Therapy (VMAT) is a cornerstone of modern radiation therapy, enabling highly conformal tumor irradiation and healthy-tissue sparing. Yet, its planning solves inverse and nested optimization for multi-leaf collimators, monitor units and dose parameters, while enforcing their consistency to ensure mechanical deliverability. Nevertheless, this process often requires repeated re-optimization when treatment configurations change, resulting in substantial planning time per patient. To address these problems, we present a diffusion-driven Learning-to-Optimize (L2O) method for"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"we present a diffusion-driven Learning-to-Optimize (L2O) method for end-to-end VMAT planning. A distribution-matching distilled diffusion model learns a clinically feasible manifold of fluence maps, enabling their one-shot generation. On top of this, an LSTM-based L2O module learns gradient update dynamics to swiftly refine fluence maps toward prescribed dose objectives during inference.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The learned fluence map manifold is clinically feasible and the LSTM module can learn stable gradient update dynamics that generalize to new patient geometries without post-hoc tuning or safety overrides.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A distilled diffusion model generates clinically feasible fluence maps for VMAT and an LSTM-based optimizer refines them to meet dose objectives, improving efficiency and deliverability on prostate cancer data.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A distilled diffusion model generates clinically feasible fluence maps in one shot for VMAT radiotherapy, then an LSTM refines them to meet dose goals.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f49234f1b01b87ac3aea4ee5fec527c225d4d30cdd8cd576e71c46befbbafdc9"},"source":{"id":"2605.13713","kind":"arxiv","version":1},"verdict":{"id":"e9c0ba67-6df9-418e-ad6c-4745c16e4a83","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:05:37.673305Z","strongest_claim":"we present a diffusion-driven Learning-to-Optimize (L2O) method for end-to-end VMAT planning. A distribution-matching distilled diffusion model learns a clinically feasible manifold of fluence maps, enabling their one-shot generation. On top of this, an LSTM-based L2O module learns gradient update dynamics to swiftly refine fluence maps toward prescribed dose objectives during inference.","one_line_summary":"A distilled diffusion model generates clinically feasible fluence maps for VMAT and an LSTM-based optimizer refines them to meet dose objectives, improving efficiency and deliverability on prostate cancer data.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The learned fluence map manifold is clinically feasible and the LSTM module can learn stable gradient update dynamics that generalize to new patient geometries without post-hoc tuning or safety overrides.","pith_extraction_headline":"A distilled diffusion model generates clinically feasible fluence maps in one shot for VMAT radiotherapy, then an LSTM refines them to meet dose goals."},"references":{"count":33,"sample":[{"doi":"","year":2026,"title":"arXiv preprint arXiv:2603.06338 (2026)","work_id":"cf2ce4c6-fee6-456e-a047-bb588ab54fae","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Medical Physics52(5), 3183–3190 (2025)","work_id":"08aa94a4-c787-4929-b67c-c7722ca2f044","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"In: Seminars in radiation oncology","work_id":"46722050-da4c-4654-839e-fda2bf79d44b","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"arXiv preprint arXiv:2305.18014 (2023)","work_id":"84d4d1c3-adeb-4381-b7eb-e520e6a52a25","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Physics in Medicine & Biology68(15), 155006 (2023)","work_id":"e185226a-2edd-447d-821d-2d81fb292fd3","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":33,"snapshot_sha256":"3f8949bcb84b1965f5667e6a79a1fcd43872493034eff55a7008b1f4cfd5627b","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"ee365135b13fbe77dcea8590b131bc8a38fd1a18f7607d1010bd6f6496317ad6"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}