S2360
Physics - Quality assurance and auditing
ESTRO 2026
identified for the 1%/1 mm dataset, with mean GPR values of 74.03 ± 6.86% and 88.80 ± 5.18%, corresponding to increasing plan quality and dosimetric robustness.
Digital Poster 2188
Preliminary Development of a Predictive Model for CyberKnife SBRT Dose Quality Assurance Based on MLC Complexity Metrics Rocco Mottareale 1 , Marcello Serra 1 , Gianluca Ametrano 1 , Valentina d'Alesio 1 , Francesca Buonanno 1 , Cecilia Arrichiello 1 , Gaetano Gagliardo 2 , Vincenzo Ravo 1 1 Radiation Oncology, Istituto Nazionale Tumori - IRCCS Fondazione "G. Pascale", Napoli, Italy. 2 Radiation Oncology, Fondazione Muto ETS, Napoli, Italy Purpose/Objective: This study aimed to develop a predictive statistical model for dose quality assurance (DQA) outcomes in Stereotactic Body Radiotherapy (SBRT) plans delivered using the CyberKnife system equipped with a Multileaf Collimator (MLC). The objective was to quantify the relationships between plan complexity metrics and measured dosimetric accuracy, supporting future automation of CyberKnife treatment plan verification. Material/Methods: A total of 30 SBRT plans were analyzed, each verified through the SRS MapCHECK system (Sun Nuclear). DQA outcomes were expressed as gamma passing rates (GPR) in accordance with AAPM TG-218 recommendations for 2%/1 mm and 1%/1 mm criteria. Complexity parameters were extracted from CyberKnife CKPlanDetails files, including: Monitor Units (MU), Modulation Complexity Score (MCS), Leaf Travel (LT), Edge Metric (EM), Plan Irregularity (PI), Plan Modulation (PM), Weighted Leaf Gap (WLG), and Small Aperture Score <10 mm (SAS10). Following data cleaning, correlation matrices were generated to assess relationships between MLC metrics and DQA indices. Multiple linear regression models were developed to predict GPR values, with Pearson’s correlation coefficient used to evaluate model performance. Additionally, unsupervised K-Means clustering, optimized using the silhouette score was applied to identify natural groupings between measured and predicted outcomes. Results: For the 2%/1 mm criterion, the regression model achieved r = 0.695, with MCS (+119.20), PI ( − 13.67), and PM (+106.25) identified as the strongest predictors. For the 1%/1 mm criterion, r = 0.757, and PI ( − 21.18), PM (+22.48), and SAS10 (+34.67) exerted the greatest influence. Both models captured the primary variability in measured DQA results, confirming that dosimetric accuracy depends strongly on plan modulation and geometric complexity. K-Means clustering revealed three clusters for the 2%/1 mm dataset, with mean GPR values of 79.15 ± 3.96%, 92.61 ± 2.99%, and 96.61 ± 4.55%. Two clusters were
Conclusion: Predictive modeling of CyberKnife SBRT DQA outcomes based on MLC complexity metrics demonstrated moderate correlations between plan modulation parameters and measured GPR values. MCS, PI, and PM emerged as key predictors of dosimetric accuracy. These findings support the integration of predictive analytics into CyberKnife QA workflows to enhance efficiency and reliability in treatment plan verification, as this activity streamlines the number of QAs to be performed—since CyberKnife QAs have the same duration as a treatment—thereby optimizing the overall workflow. Keywords: CyberKnife SBRT, DQA, MLC Complexity Metrics
Made with FlippingBook - Share PDF online