Investigation of the characteristics of multileaf collimator position errors: dose distribution- based error classification

Abstract

Background The characteristics of the multileaf collimator (MLC) position error were investigated without clinical variability and other factors affecting the error analysis. An index indicating the attributes of MLC position error was found and used for MLC position error prediction model. The dose-volume histogram (DVH) was examined to investigate the clinical relationship.Methods: The dose distribution was investigated using the gamma index, structural similarity (SSIM) index, and dosiomics index. The cases from the American Association of Physicists in Medicine Task Group 119 were planned, and systematic and random MLC position errors were simulated. All error-free and error datasets were generated in the treatment plan system. The indices were obtained from distribution maps, and then statistically significant indices were selected. An MLC position error prediction model was developed using the selected indices and logistic regression method. The final model was determined when all values of the area under the curve (AUC), accuracy, precision, sensitivity, and specificity were higher than 0.8 (p<0.05). DVH relative percentage difference between the error-free and error datasets was examined to investigate clinical relations.Results: Statistically common significant indices were found, GLCM_Energy in Class-I and Class-III and GLRLM_LRHGE in Class-II. The final model was developed using indices that satisfied the statistical criteria. The number of finalized univariate predictive models was five in Class-I and Class-II and four in Class-III. Seven multivariate predictive models were finalized. The DVH relative percentage difference between the error-free and error dataset almost linearly increased as systematic error increased. In the case of random errors, the tendency of the DVH relative percentage difference was dependent on the structure’s location.Conclusion: Our study highlights three novel vital results. First, the common significant dosiomics indices (GLCM Energy and GLRLM_LRHGE) can characterize the MLC position error. Second, the finalized logistic regression model for MLC position error prediction showed excellent performance with AUC > 0.9. Third, the results of DVH were related to dosiomics analysis in that it reflects the characteristics of the MLC position error, and it was shown that dosiomics analysis could provide important information on localized dose distribution differences in addition to DVH information.