EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Autonomous radiotherapy planning via agentic orchestration using a multimodal TPS-integrated compound AI platform

Austen Matthew Maniscalco iD, Yang Kyun Park iD, Sean J Domal iD, Mu-Han Lin iD, Sanda Harabagiu iD, Steve B Jiang iD, Dan Nguyen iD

Machine Learning: Health · 31.12.2026 · 2 (2) · 025005

DOI: 10.1088/3049-477x/ae7978 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Machine Learning: Health, online erschienen am 15.07.2026. Verfasst von Austen Matthew Maniscalco und 6 weiteren Autor:innen. Beteiligte Einrichtung: Medical Artificial Intelligence and Automation (MAIA) Lab, UT Southwestern Medical Center, Dallas, TX, United States of America. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 66 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: NCI NIH HHS, National Institutes of Health.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Purpose

Radiation therapy (RT) treatment planning requires iterative, multi-day optimization workflows in which subjective planning strategies produce inter-planner variability in plan quality. Existing computational approaches automate isolated aspects of this workflow, yet none orchestrates an end-to-end pipeline from physician directive to deliverable plan. We developed a compound artificial intelligence (AI) platform for fully autonomous RT treatment planning that combines multi-agent large language model (LLM) orchestration with directive-conditioned three-dimensional (3D) dose prediction, natively integrated with a commercial treatment planning system (TPS).

Methods

Seven specialized agents navigated the multi-objective optimization landscape through structured clinical reasoning, iteratively analyzing dose-volume histogram (DVH) metrics and spatial dose patterns, formulating trade-off strategies, and executing validated modifications through the TPS across five fully autonomous iterations per case. A directive-conditioned 3D dose prediction model supplied patient-specific DVH values from which initial optimization objectives were autonomously derived, eliminating the need for curated templates or manual initialization. A retrieval-augmented generation (RAG) system encoded institutional knowledge into the planning workflow. We evaluated 60 retrospective cases across brain, lung, and prostate sites, with 10 intensity-modulated RT (IMRT) and 10 volumetric modulated arc therapy (VMAT) plans per site spanning 20.0-79.2 Gy in 3-44 fractions, scored by the proportion of dosimetric criteria satisfied.

Results

Across all 60 cases, AI plans achieved 89.8 ± 9.4% of dosimetric criteria versus 85.2 ± 10.8% for clinical reference plans ( p < 0.001 ). IMRT plans improved in 25 of 30 cases with none worsened (94.1 ± 6.7% vs 84.3 ± 11.8%, p < 0.001 ); VMAT plans showed no significant difference (85.6 ± 9.9% vs 86.1 ± 9.7%, p = 0.770 ). Each plan iteration completed in 20.2 ± 12.7 min, of which agent reasoning consumed 5.2 ± 1.7 min ( 114,429 ± 11,798 tokens, 0.43 ± 0.04 ).

Conclusions

These results established the feasibility of end-to-end, fully autonomous, universal RT treatment planning through compound AI. Integrating dose prediction as an agent-invoked tool for objective initialization resolved the dependency on curated templates and manual specification that constrained prior LLM-based planning systems.

Schlagwörter

Stichwörter der Autor:innen

LLM agentsagentic RAGautonomous treatment planningcompound AI systemsdose predictionmulti-agent orchestrationradiation therapy

Autor:innen und Institutionen

  • Medical Artificial Intelligence and Automation (MAIA) Lab, UT Southwestern Medical Center, Dallas, TX, United States of America.
  • Department of Radiation Oncology, UT Southwestern Medical Center, Dallas, TX, United States of America.
  • University of Texas at Dallas, Richardson, TX, United States of America.

ORCID-Profile: Austen Matthew Maniscalco, Yang Kyun Park, Sean J Domal, Mu-Han Lin, Sanda Harabagiu, Steve B Jiang, Dan Nguyen

Förderung

  • NCI NIH HHS
  • National Institutes of Health

Zitieren

APA

Matthew Maniscalco, A., Kyun Park, Y., Domal, S. J., Lin, M. H., Harabagiu, S., Jiang, S. B., & Nguyen, D. (2026). Autonomous radiotherapy planning via agentic orchestration using a multimodal TPS-integrated compound AI platform. Machine Learning: Health, 2(2), 025005. https://doi.org/10.1088/3049-477x/ae7978

Vancouver

Matthew Maniscalco A, Kyun Park Y, Domal SJ, Lin MH, Harabagiu S, Jiang SB, et al. Autonomous radiotherapy planning via agentic orchestration using a multimodal TPS-integrated compound AI platform. Machine Learning: Health. 2026;2(2):025005. doi:10.1088/3049-477x/ae7978

BibTeX
@article{MatthewManiscalco2026,
  author = {Matthew Maniscalco, Austen and Kyun Park, Yang and Domal, Sean J and Lin, Mu-Han and Harabagiu, Sanda and Jiang, Steve B and Nguyen, Dan},
  title = {Autonomous radiotherapy planning via agentic orchestration using a multimodal TPS-integrated compound AI platform},
  journal = {Machine Learning: Health},
  year = {2026},
  volume = {2},
  number = {2},
  pages = {025005},
  doi = {10.1088/3049-477x/ae7978}
}
Kostenlos starten →