PROJECT SUMMARY
Recently, the development of MR-LINACs has made high-quality online adaptative radiotherapy a clinical
reality to account for the daily anatomical variations to preserve the treatment quality. MR-LINACs, combining
modern radiotherapy linear accelerators (LINACs) with on-board magnetic resonance imaging (MRI), offer
excellent soft-tissue contrast to allow accurate organ and tumor segmentation to precisely capture the daily
anatomical changes of each patient. Coupled with advanced adaptive treatment planning systems, MR-LINAC
is the ideal platform for online adaptive radiotherapy and will bring cancer radiotherapy to a new level of
precision and personalization. However, this new format of radiotherapy also comes with new challenges for
patient safety and plan quality checks that cannot be satisfactorily addressed with traditional quality assurance
(QA) tools: 1) With the patient lying on the treatment couch waiting for the treatment to start, there is mounting
pressure on the team to move through the workflow as fast as possible, which may increase the likelihood of
making mistakes and thus an effective QA procedure is even more important. 2) Each adapted plan warrants a
new QA process, adding substantial burdens to an already extremely time-constrained process. A QA process
with high efficiency is needed. 3) Conventional QA procedures are quite complex, involving inputs from many
stakeholders, and thus are human-power demanding and error-prone. An automatic QA procedure requiring
minimal human interventions and communications is highly desired. 4) In addition to checking the quality of the
adapted segmentation and treatment plan, it is also crucial for a QA procedure to ensure their consistency with
the physician’s intentions/preferences in the original plan. 5) A QA tool that is able to predict the plan
deliverability prior to treatments, without actually irradiating the patients, is needed for online adaptive
radiotherapy. The overarching goal of this project is to develop an Artificial Intelligence (AI)-based QA system
to address these urgent unmet clinical needs for MR-LINAC online adaptive radiotherapy, with four main
components to: 1) intelligently assess the quality of the adapted target and organ-at-risk segmentations and
their consistency with those in the original plan; 2) intelligently assess the quality of the adapted plan and its
consistency with the original plan; 3) efficiently perform 2nd dose check with an AI-based near real-time
independent dose engine; and 4) predict the measurement-based QA results of plan deliverability using prior
knowledge and new adapted plan information. We have two Specific Aims: 1) System development, including
data acquisition for AI model training, and development of four AI models; and 2) System translation and
validation at multiple institutions, including developing transfer learning algorithm and package for automated
model commissioning; and translation, fine-tuning and evaluation of the developed AI systems. The successful
conduct of the proposed project will result in the first intelligent, efficient, reliable, and independent QA system
to facilitate unleashing the full potential of MR-LINAC online adaptive radiotherapy to advance cancer care.