On-treatment Adaptation of Head and Neck Cancer Dosimetry Plans Using Novel Cone-beam CT and AI-driven Software

This study focuses on imaging and radiotherapy treatment plan modification for head \& neck cancer patients using a high-performance cone beam CT (CBCT) image guidance system (HyperSight) and AI-driven adaptive planning software. The HyperSight imaging system is a forthcoming component of a radiation therapy (RT) treatment platform, called Ethos (Varian Medical Systems).

RT is used in the treatment of approximately 75% of all head and neck (H\&N) cancers with either curative or palliative intent. External beam radiation therapy is typically delivered over 5-7 weeks in 25-35 daily treatments (or "fractions"). Image guided radiation therapy (IGRT) is now a standard of care for external beam radiotherapy. During an IGRT treatment session, a CBCT imaging system built into the treatment unit is used to acquire 3D images of the patient's anatomy immediately prior to, and, if required, at several time points during, the daily delivery of radiation. The CBCT image data are used by the treatment team to carefully align the patient to a treatment planning fan-beam CT (FBCT) images that was acquired 1-2 weeks prior to the start of treatment, and which was used to generate the dosimetry plan that determines the daily radiation delivery. Verification of the patient position and alignment with the treatment planning images increases the precision of radiation delivery.

Daily imaging is particularly important for accurate setup of patients treated for head and neck (H\&N) cancer, as these patients often experience weight loss and target volume changes throughout the course of treatment. These volume changes can alter the expected dose distribution, as organs at risk (OAR) migrate to high dose regions or target volumes develop inhomogeneities. When volume changes are perceived by the treatment team to have a significant impact on radiation delivery, patients may be sent for a second planning FBCT and have a new treatment plan created for the remainder of their treatment. This process is referred to as offline replanning or offline adaptation. The radiation oncologist must decide whether to pause treatment or continue with the original treatment plan until a new plan is created. In either case, it can take several days between the point when an issue is identified and when a new plan is ready to be delivered. Offline replanning is not a regularly scheduled part of the clinical workflow and creates disruption of routine imaging, contouring and planning tasks for other patients. These tasks must be moved aside to free the necessary resources to create a new plan for the patient who is already receiving treatment and to permit the treatment team to start delivering the new plan as soon as possible.

The CBCT images acquired with the high-performance HyperSight CBCT imaging system have demonstrated Hounsfield Unit accuracy sufficient for accurate dosimetry calculations in phantoms. These images can therefore potentially be used to plan, or re-plan a radiation treatment delivery. The ability to use imaging that is acquired as part of the daily treatment delivery for re-planning could reduce the need to send patients for additional imaging and could also reduce the time and resources required to re-plan treatment delivery mid-course. This is of great importance for a cohort of patients already known to be vulnerable to the need for replanning.

In addition to the challenges involved in mid-treatment re-planning, another issue commonly encountered in H\&N patients is the presence of imaging artifacts from high-density metal objects in their oral cavity such as dental fillings and implants. These artifacts can cause difficulties in visualization of target volumes and OARs on planning CT as well as perturbation of dose calculation, particularly for cancers in the oral cavity. Metal artifacts can also cause a significant reduction in CBCT image quality, limiting the usefulness of IGRT for patient positioning.

The HyperSight CBCT imaging system incorporates several hardware and software changes that have the potential to provide higher quality CBCT images including improved visualization of volume changes and improved metal artifact reduction.

With minimal disruption for participating patients, this study will enable a comparison of (i) the patient's treatment planning FBCT and (ii) the patient's conventional CBCT on an existing treatment unit (Truebeam CBCT) with (iii) patient CBCT images on the HyperSight system. Image quality of the high performance CBCT will thereby be compared to both a best-case standard (FBCT) and the conventional CBCT for IGRT (Truebeam CBCT) to identify potential improvements in the HyperSight CBCT images. Image quality will be quantified in terms of contrast-to-noise ratio (CNR), resolution, Hounsfield Unit (HU) accuracy and artifact reduction with a focus on anatomy germane to the image guidance task. A qualitative assessment of the different image datasets will also be performed through feedback from multiple observers (radiation oncologists and radiation therapists). Radiation doses to key targets and organs at risk will be calculated from the high performance CBCTs and compared to the doses calculated in the subject's treatment plan to determine the suitability of the CBCT images for RT dosimetric planning. Various re-planning scenarios will also be examined to determine potential increases in efficiency from using the HyperSight CBCT imaging system for re-planning.

Image sets (i) and (ii) above are already acquired as part of standard-of-care for radiation treatment. Two additional CBCTs, taken on the day of the patient's treatment planning FBCT and again on the day the patients receive their 21st treatment fraction (approximately two-thirds through treatment) will be acquired from each subject on the HyperSight CBCT imaging system for this study. Not every patient will require mid-treatment re-planning but based on prior institutional data, if re-planning is required, the indications for re-planning should be evident in most of that population by the time 2/3 of their treatment has been delivered. If the treatment team determines that a patient requires a second planning FBCT during treatment, the patient will be scheduled for a third HyperSight CBCT on the same day that they receive their second planning FBCT.

Radiation treatment planning and delivery will not be altered due to patients' participation in this study. Patients will receive radiotherapy according to current departmental procedures and on the department's existing linear accelerator platforms. The decision to obtain a second planning CT and to generate a new mid-treatment plan will not be affected by a subject's participation in the study.

The HyperSight CBCTs that are acquired in this study will allow the examination of several re-planning strategies. The process of mid-treatment re-planning can be simulated for all subjects using the HyperSight CBCT acquired at fraction 21. Timing and workflow efficiencies can be evaluated in these simulated re-planning processes and compared to institutional norms (e.g., average time from identification of the need for re-planning to the start of delivery of a new treatment plan). In the subset of study subjects who require re-planning as part of their treatment, the HyperSight CBCTs acquired on the day of the second FBCT will also be used to generate a treatment plan and the dosimetry of the HyperSight CBCT-based re-plan will be compared to the dosimetry of the mid-treatment re-plans generated for these subjects using the current offline re-planning methods.

In addition to the simulation of current offline re-planning methods using the HyperSight CBCT images, this study will evaluate the online adaptive workflow that is available through the Ethos Radiotherapy System. Online adaptive workflow leverages AI-based anatomical structure contouring and plan optimization tools to adjust the daily radiation treatment delivery. Instead of delivering the same treatment in every fraction, the treatment plan defined prior to the start of treatment is optimized based on the anatomy captured in the CBCT acquired at the start of the daily fraction. Independent of the patient's clinical treatment plan, the AI-driven adaptive planning component of Ethos will be used to simulate online adaptation of the treatment plan. This study will evaluate the potential dosimetric benefits of daily online adaptive planning in Head and Neck cancer and the potential for online adaptation to reduce the need for offline re-planning. The time and resources required to implement an online adaptive workflow will be compared to standard IGRT delivery.