Decipher dose accumulation for online adaptive radiotherapy

The National Cancer Institute awarded The University of Texas MD Anderson Cancer Center a cooperative agreement totaling $425,410 on September 1, 2026, under the Cancer Treatment Research program (CFDA 93.395) to develop artificial intelligence-based deformable image registration tools for dose accumulation in online adaptive radiotherapy.

Online adaptive radiotherapy enables real-time adjustments to radiation treatment plans during delivery by re-planning based on newly acquired images; however, current commercial systems on MR-guided platforms (Elekta Unity MR-Linac) and cone beam CT-guided platforms (Varian ETHOS) do not accurately estimate or display cumulative doses to disease targets or organs at risk across multiple treatment fractions. The project addresses this unmet clinical need by creating AI-driven tools to automate dose accumulation, which depends on precise deformable image registration across varying patient anatomy and tumor changes including shrinkage or disappearance. Accurate dose quantification is essential for clinicians to verify that dosimetric criteria are met over the course of treatment and to guide online adaptive radiotherapy toward therapeutic goals.

The award runs through August 31, 2029, with place of performance in Texas. The $425,410 obligated amount represents the initial funding for this three-year effort to close a significant gap in commercial online adaptive radiotherapy platforms, particularly for MR-Linac systems increasingly used in clinical practice.

Abstract

Online adaptive radiotherapy (art) allows for real-time adjustments of radiation plans to account for daily anatomical variations by RE-planning while the patient is on the treatment table, using newly acquired images. While this process can be resource-intensive, innovations in workflows by various teams have made art a reality in clinical practice. These workflows build on commercial online art platforms including the elekta unity MR-linac, which uses MR-guided technology, and the varian ethos, which is a cone beam CT (cbct)-guided platform. Despite progress in implementing online art, there is untapped potential. Current systems do not accurately estimate or display cumulative doses to disease targets or specific organs at risk over the course of treatment. The ability to accurately quantify delivered dose is important for clinicians to evaluate whether dosimetric criteria are being met in an aggregation over multiple fractions, as well as to guide online art to achieve ultimate therapeutic goals. Accurate dose accumulation depends on precise deformable image registration, but this is challenging due to anatomical variations. In addition, quick evaluation of deformation accuracy for online art remains impractical, and the appropriate dose accumulation approach for tumor shrinkage or disappearance is unknown. Despite ongoing research on dose accumulation, few solutions have been implemented in commercial systems, particularly for MR-linac online art platforms. This GAP represents a significant unmet clinical need, given the increasing use of online art in clinics. The primary goal of this project is to develop artificial intelligence (AI)-based deformable image registration tools for effective and automated dose accumulation, addressing urgent unmet clinical needs in online art. Our central hypothesis is that these AI-based deformable registration tools will enable accurate and automated daily dose accumulation, revealing the dosimetric impact of daily anatomy changes. This information will facilitate informed online adaptive decision-making, reducing normal tissue dose while achieving the target dose. We will accomplish the objective of this proposal by pursuing the following specific aims: 1) algorithm development and prototyping. In this aim, we will develop an AI-based deformable registration algorithm specifically for dose accumulation, AI-based quick deformation error detection tool, and an AI-guided interactive deformation editing tool. 2) implementation and validation of dose accumulation for online art. In this aim, we will implement and evaluation dose accumulation tools to inform online adaptive planning and decision-making for both MR-linac and ethos systems. Successfully completion of this project will result in the first intelligent, efficient, and reliable automated dose accumulation system to facilitate online art to advance cancer care. Ultimately, this work will unleash the power of online art to achieve better precision and personalized radiation oncology.