Application of error classification model using indices based on dose distribution for characteristics evaluation of...

Area of Science:

• Medical Physics
• Radiotherapy Physics
• Image Analysis in Radiation Oncology

Background:

• Accurate radiation delivery relies on precise multileaf collimator (MLC) positioning.
• MLC position errors can significantly impact dose distribution and treatment efficacy.
• Advanced analysis techniques are needed to characterize these errors effectively.

Purpose of the Study:

• To evaluate specific characteristics of MLC position errors correlated with dose distribution indices.
• To investigate the utility of gamma, structural similarity, and dosiomics indices for error characterization.
• To develop a predictive model for MLC position errors using dosiomics features.

Main Methods:

• Simulated systematic and random MLC position errors using American Association of Physicists in Medicine Task Group 119 plans.
• Analyzed dose distributions using gamma, structural similarity, and dosiomics indices.
• Developed and validated multivariate predictive models, including logistic regression, assessing performance metrics (AUC, accuracy, precision, sensitivity, specificity).
• Examined dose-volume histogram (DVH) differences and correlated them with dosiomics findings.

Main Results:

• Seven multivariate predictive models were finalized, with significant dosiomics indices (GLCM Energy, GLRLM_LRHGE) identified as key characterizers of MLC position errors.
• The logistic regression model achieved excellent performance for MLC position error prediction (AUC > 0.9).
• Dose-volume histogram (DVH) analysis results correlated with dosiomics, reflecting MLC position error characteristics and highlighting localized dose differences.

Conclusions:

• Dosiomics analysis, particularly using GLCM Energy and GLRLM_LRHGE, effectively characterizes multileaf collimator (MLC) position errors.
• A predictive model based on dosiomics demonstrates high accuracy in identifying MLC position errors.
• Dosiomics offers complementary information to DVH, enhancing the understanding of localized dose distribution changes due to MLC errors in radiotherapy.