Research in Action: Abstract Highlights From AAPM | COMP 2026

At the 2026 Joint AAPM | COMP Annual Meeting in Vancouver, British Columbia, researchers presented new work exploring how intelligent automation, adaptive assessment, Monte Carlo dose calculation, and log file analysis can support the demands of modern radiation oncology.

These studies evaluated Radformation solutions in settings ranging from phantom validation to real-world implementation. Together, they offer a look at how departments are leveraging smart tools to address long-standing challenges while building the clinical evidence to support new approaches to care.

Detecting change and informing adaptive decisions

Phantom-based validation evaluating accuracy and consistency
Over the course of treatment, patient anatomy can change. While we have systems designed to account for some of those changes (structure margins, daily imaging), sometimes change is clinically meaningful enough to warrant replanning. But these decisions to adapt are often subjective. Two studies explored how ChartCheck Adaptive can help bridge that gap by turning daily imaging into quantitative information clinicians can use to evaluate changes throughout treatment.

At the University of Kansas, Kenny Guida and colleagues evaluated ChartCheck Adaptive using phantom studies designed to test the system under both static and changing anatomical conditions. VMAT prostate and C-shape plans were delivered to solid water, while a head-and-neck RANDO phantom with a target made of putty was modified during the simulated treatment course to introduce anatomical changes. ChartCheck Adaptive automatically imported treatment data, generated synthetic CTs, performed Monte Carlo dose calculations, and compared delivered dose with the initial plan.

For unchanged prostate anatomy, results show excellent dose agreement between planned and delivered doses (within 0.6% for all fractions), while the H&N study demonstrated volume fidelity to within 2% on average, even after reshaping bolus material to simulate changes in setup. This demonstrates how daily dose and DVH information can help clinicians connect changes in dose coverage or OAR sparing with anatomical changes and identify cases that may warrant further evaluation for replanning.

Phantom studies with the ChartCheck Adaptive platform demonstrated consistency of volume and dose metrics in the absence of change, showing volume fidelity to within 2% on average even when bolus material positioning was manipulated.

The metrics that stand out in cases that require replanning
Researchers at UC San Diego investigated ChartCheck Adaptive using a cohort of head and neck patients. Irena Dragojević, PhD, and her team retrospectively evaluated 16 patients, half of whom underwent re-simulation for replan assessment during treatment and half of whom did not. ChartCheck Adaptive tracked changes in intermediate- and low-risk CTVs throughout treatment, as well as the left and right parotids.

For patients that needed re-simulation, volume changes greater than 10% were observed consistently as treatment progressed, particularly after 15 fractions. In contrast, patients not selected for re-simulation remained within the 10% threshold throughout treatment. Following replanning, volume change trends became substantially flatter when measured relative to the re-simulation fraction. The results correlate changes in key dose metrics to a clinical decision to replan.

Together, these studies illustrate a progression from detecting anatomical and dosimetric change to using data to help identify when intervention may be appropriate. Instead of relying exclusively on subjective visual review of daily images, automated tracking can provide clinicians with objective metrics to support adaptive decision-making.

RadMonteCarlo validation: from beam data to end-to-end delivery
Independent dose calculation depends on confidence in the underlying calculation engine. Two studies from Washington University in St. Louis approached validation of RadMonteCarlo from complementary perspectives, first comparing calculations against commissioned beam measurements and then against independent end-to-end IROC phantom data.

Open-field measurement comparisons
Phillip Wall and colleagues performed a comprehensive open-field dosimetric evaluation of RadMonteCarlo using chamber-measured photon beam data. Three-dimensional Monte Carlo simulations were compared with measured data using MPPG 5.a dose-difference and distance-to-agreement criteria.

Across the evaluated beam energies and field sizes, agreement was strong, with all points passing on PDD comparisons across all energies and beam sizes. In aggregate, dose profiles measured across all depths and field sizes passed 99%+ of points, with disagreements localized in low-dose, out-of-field, or tail regions.

Comparing RadMonteCarlo dose to IROC phantom film profiles
A second study led by Guilherme Ferreira approached validation through an independent end-to-end test. Using 12 anthropomorphic IROC phantoms representing brain, head-and-neck, and lung treatments, researchers compared RadMonteCarlo calculations with IROC film profiles and TLD data.

All film profile comparisons met standard IROC 7%/5 mm gamma criteria, and all TLD comparisons fell within the IROC ±7% tolerance. All but one case passed 90% of points using tighter criteria, which revealed greater sensitivity in challenging brain cases involving small fields and steep dose gradients.

Box plot showing the percent difference between RadMonteCarlo-calculated TLD dose values and IROC-reported measurements. Across all sites and phantom types, differences were within 4.4%.

Taken together, the studies evaluate RadMonteCarlo with two unique methods, showing agreement with commissioned beam measurements and performance within realistic, independently measured treatment scenarios.

Can log file analysis detect introduced errors?
Jonathan Pence and colleagues at Washington University investigated this very question using trajectory log files and RadMonteCarlo. The researchers collected treatment delivery logs from 20 patient treatments and introduced simulated MLC deviations including opening, shift, skew, and backlash errors. Plans were recalculated with RadMonteCarlo to evaluate the dosimetric effects of those deviations.

The analysis showed some level of sensitivity to all the introduced errors, with systemic MLC opening and shift producing the largest average decreases in 3D gamma pass rate. The study suggests that daily trajectory log analysis could provide an additional way to identify divergent delivery parameters during a treatment course, including changes potentially associated with mechanical wear.

Rather than viewing PSQA solely as a pretreatment intervention, this approach points toward a more continuous, on-treatment verification model using information automatically generated with each fraction to help monitor delivery consistency.

Difference in 3D gamma pass rate between original log files and modified MLC leaf positions, showing higher detectability for errors such as shifts or openings versus skew.

Evaluating log file analysis for IMRT/VMAT QA
Nicholas Harvey at Hofstra University evaluated log file treatment delivery data from a different angle: whether log file analysis could serve as an efficient approach to patient-specific QA for IMRT and VMAT plans.

Harvey’s team evaluated 22 plans delivered on two Varian TrueBeam systems. Trajectory logs containing information such as MLC position, gantry motion, and dose rate were automatically imported, and RadMonteCarlo calculated delivered dose on the patient CT. Gamma analysis was then performed using TG-218 criteria for both the volumetric dose matrix and high-dose PTV.

The study found good overall agreement between RadMonteCarlo and AAA, with some notable differences in regions of heterogeneous anatomy, air-tissue interfaces, implanted hardware, or buildup consistent with the inherent differences in the dose calculation algorithms. For these cases, relaxed gamma criteria were used for clinical acceptability. The researchers concluded that this approach can provide a robust, efficient, and clinically viable approach to PSQA.

As stated in the presented work, “[The log file analysis] workflow reduces the need for phantom setup and physicist labor, thereby improving efficiency without compromising plan validation.”

Scaling institution-specific automation with Memorial Sloan Kettering
Intelligent automation tools can reduce repetitive work, but large departments often have another challenge: how to retain highly specialized, institution-specific checks while benefiting from a standardized commercial platform.

Sean Berry and colleagues at Memorial Sloan Kettering Cancer Center (MSK)—in collaboration with Radformation—presented an approach for integrating MSK-developed automated plan checks into ClearCheck while preserving the institution's ability to develop and modify its own custom checks.

Diagram illustrating the interaction between ClearCheck and MSK’s in-house Plan Check Tool. JSON format is used to communicate between the two systems, taking patient information from ClearCheck to a custom script, then writing back to ClearCheck for display.

The implementation was released across seven campuses and approximately 200 users over a six-month period. As commercially available checks replaced redundant institution-maintained checks, MSK reduced its custom checks from 105 to 94 and developed 12 additional custom checks. Alongside ClearCheck, the combined environment ultimately included 150 automated checks. All custom scripts are combined with ClearCheck outputs in a comprehensive plan evaluation report.

The study highlights a model in which commercial and institution-developed automation can coexist to produce rigorous and cohesive treatment plan evaluations. Rather than requiring clinics to choose between standardization and customization, the approach allowed MSK to shift internal development resources toward checks specific to its clinical needs. As a bonus that benefits others, this new feature is now available to all in ClearCheck v2.7.

From research to clinical workflow
From recognizing meaningful anatomical changes during treatment, to independently validating dose calculations, to monitoring treatment delivery and scaling automated plan review, these studies demonstrate how clinical partners are investigating new ways to bring objective data into workflows that have traditionally required significant manual effort.

We’re excited to see institutions independently studying these applications, identifying both their opportunities and limitations, and continuing to build the evidence needed to translate automation into practical clinical workflows.

Want to see what intelligent automation from Radformation can do for your department? Contact us to schedule a personalized demo today.