1 Abstract

Background: Volumetric Modulated Arc Therapy (VMAT) is the state of the art treatment delivery method for prostate cancer patients, enabling the dose to conform tightly around the target volume and spare the surrounding healthy tissues. However, VMAT plans are inherently complex, affecting treatment deliverability, radiation-induced carcinogenesis, and mechanical strain on the linear accelerator. Moreover, they require laborious quality assurance procedures to ensure the safety of the patients. By determining which plan features are related to complexity, we may gain insights into how to regulate the treatment planning process to keep complexity under control.

Objective: To compare multiple complexity metrics described in the literature and investigate which VMAT plans’ characteristics relate to complexity in prostate cancer patients undergoing radical external beam radiation treatment.

Methods: We calculated circumference over area, edge metric, equivalent square field, leaf travel, leaf travel modulation complexity score for VMAT (LTMCSV), mean-field area, modulation complexity score (standard and VMAT variant), plan irregularity, and short aperture score. These were studied with principal component (PCA) and mutual information analysis (MIA). LTMCSV was selected for subsequent regression analysis. The plan-related variables that were evaluated as potential predictors of complexity included the total number of monitor units, number of arcs, radiation field size, physician, dosimetrist, treatment planning system (TPS) version, conformation number (CN), and high-dose planning treatment volume (PTV). Also, the dose-volume constraints, bladder \(V_{65}\), \(V_{70}\), rectum \(V_{50}\), \(V_{60}\), and \(V_{70}\) were considered. Linear and logistic regression analyses were performed by treating LTMCSV as a continuous and dichotomous variable, respectively. In the former case, model selection was done by univariate and multivariate analysis, forward and backward selection. In the latter case, feature selection was implemented via LASSO regularization and k-fold cross-validation, whereas the final evaluation was performed with ROC analysis and confusion matrices.

Results: A total of 217 VMAT prostate plans were exported from the oncology information system of Papageorgiou General Hospital radiation oncology department. For every plan, the complexities of all the studied indices were calculated. PCA analysis revealed that complexity metrics are correlated, forming three distinct clusters, each subsuming a different aspect of a plan’s complexity. Three principal components were able to explain 96.2% of the variance. Likewise, MIA confirmed the complexity metrics’ inherent interdependence, although there were complexity pairs with minimal MI. In univariate analysis, the number of arcs, field size, dosimetrist, TPS version, high-dose PTV, bladder \(V_{65}\), and rectum \(V_{50}\) were statistically significant determinants of complexity. In multivariate analysis, the number of arcs (p<0.001), increasing field size (p<0.001), certain physicians (p=0.041 and p=0.027), recent TPS version (p<0.001), high-dose PTV (p=0.02), high CN (p=0.036), rectum’s \(V_{50}\) (p<0.001), and \(V_{60}\) (p=0.001) retained their statistical significance. In logistic regression, LASSO regularization invoked sparsity by driving some of the variables’ coefficients to zero. The number of arcs, field size, physician, dosimetrist, TPS version, CN, high-dose PTV, bladder \(V_{70}\), and rectum \(V_{70}\) had non-zero coefficients.

Conclusions: There is an abundance of complexity indices for IMRT/VMAT in the literature. Both PCA and MIA analyses suggest the existence of substantial overlap among the metrics. However, this overlap is not entirely reducible through dimensionality reduction techniques, implying that there also exists some complementarity. Prediction of VMAT plan complexity and classification of plans into high- or low- complexity by clinical and dosimetric features is feasible through both linear and logistic regression analyses.