To our knowledge, this is the largest real world study investigating the clinical experience using daily intelligence-assisted online adaptive radiotherapy on cone-beam CT for head and neck cancer available to date including 770 adaptive sessions. We found that daily ART on CBCT improved target coverage for high, intermediate and low-dose PTVs and reduced hot spots in the high dose PTV. In addition, selected OARs were spared with adaptation, although for some OARs this only becomes evident as the patient progresses with treatment (see Fig. 2). Our study cohort predominantly consisted of patients with locally advanced head and neck cancer. Despite the complexity of their cases, we observed excellent local control with only one local failure. Furthermore, our patients experienced favorable toxicity profiles, with few grade 3 toxicities and no occurrences of grade 4 or 5 toxicities. While we have a short median follow up, our initial outcomes appear to compare favorably to other series in terms of local control rates and acute and initial chronic toxicities [10,11,12].

Adaptive radiotherapy can be based either on functionally-based response using biological imaging such as PET or on anatomical imaging such as magnetic resonance imaging (MRI) or CT [13]. Functional-response based radiation holds promise to improve the therapeutic index through the ability to change dose based on functional imaging or based on response (either via dose escalation or de-escalation) and is currently being evaluated in several prospective studies. Recently published Phase 2 dose escalation study to hypoxic subvolumes using dynamic [18F] fluoromisonidazole PET-CT demonstrated the prognostic ability of such functional imaging whereby dose escalation resulted in a non-significant improvement in local control of 25% [14]. Currently several MRI adaptive studies are underway [15] using both functional and anatomic data. However, such strategies are currently only being implemented with intra-treatment imaging done on a weekly or sporadic basis. Prior to the advent of the CBCT adaptive platform performed in our study, anatomic-based adaptation based on spatial changes throughout the treatment course has been performed on ad-hoc basis or on systematic basis. Retrospective studies of ad-hoc adaptive radiotherapy have demonstrated dosimetric and potential oncologic benefits [5, 6]. Recently, a prospective Phase 3 study on weekly systematic replanning failed to demonstrate improved clinical outcomes compared to non-adapted radiotherapy other than mean parotid excretory function [7]. Our study is one of the first clinical experiences utilizing daily adaptive radiotherapy, a method which has not yet been checked prospectively.

CBCT is widely utilized for daily treatment to correct for setup errors between treatment days [15]. Automated methods for clinical re-planning on each CBCT are now clinically available [16] and was utilized in our study. CBCTs were deemed good enough quality to check the OARs and PTVs and re-contour when necessary. In line with other studies, our analysis demonstrated a statistically significant reduction in target volumes PTV59.5 and PTV56, as well as in the parotid glands, over the course of treatment [17, 18]. While a reduction in PTV70 volume was also observed, it did not reach statistical significance. Importantly, our intention was to preserve the initial gross disease and therefore the observed changes in intermediate and low dose PTVs were more likely attributed, in part, to factors such as weight loss. In our study we were only able to detect a slight dosimetric benefit in the maximum doses to the spine and brainstem. This is likely due to the high prioritization of these serial structures in treatment planning which are carried over with online re-optimization during adaptation as well as the fact that we maintained the volume of the high dose PTV. While most OARs demonstrated improved sparing, these did not reach statistical significance. However, several OARs did demonstrate a statistically significant change per additional session which can be visually appreciated in Fig. 2. We found that one of the significant advantages of utilizing online ART on CBCT is its capacity to provide insights into daily changes in target volumes and organ-at-risk (OAR) structures, enabling the clinical team to comprehend cumulative dose variations throughout the treatment course. This is a departure from traditional assumptions based on contouring and plans established at the time of simulation.

The CBCT workflow introduces additional uncertainty to treatment dose calculation that needs to be considered. The CBCT quality on this system is not considered adequate for direct dose calculation. Therefore, a synthetic CT is generated by the system for dose calculation purposes. The synthetic CT is built using a deformable registration of the planning CT to match the anatomy presented in the daily CBCT. The registration software utilizes regularization rules to ensure that the deformations are anatomically reasonable. For example, voxels cannot cross over each other in the deformation. However, this can lead to inaccuracies in the synthetic CT. For example, Hakansson et al. demonstrated that air cavities which developed over the course of treatment due to shrinking tumor volumes are not reflected in the synthetic CT [19]. Their study shows that these errors in the synthetic CT lead to small uncertainties. Future studies are underway at our institution to further evaluate the impact of synthetic CT inaccuracies.

We acknowledge several limitations in our study. This includes its retrospective nature and the single-institutional design. While we treated a diverse group of head and neck cancer patients, the sample size may have limited our ability to detect more substantial dosimetric benefits, particularly within specific patient subsets. Furthermore, the absence of mandatory additional CBCT scans prior to treatment delivery following adaptation means that potential shifts in patient positioning during adaptation may not have been consistently addressed for some of the fractions. In addition, our analysis only evaluated PTV coverage, although CTV coverage may give a better indication of the value of daily adaptive treatment. In the broader context of head and neck cancer treatment, our study underscores the feasibility of daily adaptive radiotherapy to improve target coverage and potentially improve OARs. However, this advantage does come at the cost of increased total treatment time, affecting both patient comfort and overall resource utilization. Moreover, given that such improvements were minimal and not seen in the majority of the OARs, we question whether utilization of daily ART in all curative intent treatments for head and neck cancer patients will translate into improved outcomes.

Further prospective studies and ongoing research are essential to identify the patient cohorts that may derive the most benefit from daily online ART. Ongoing studies are exploring whether daily online ART may allow for reduced PTV margins thereby reducing long-term toxicity and improving quality of life parameters while maintaining local control [20]. In conclusion, our study represents a substantial step demonstrating the feasibility of daily adaptive radiotherapy for head and neck cancer. It demonstrates the potential to improve radiotherapy, offering improved daily target coverage, which may ultimately improve outcomes for patients facing this challenging disease.