Classification of PSQA outcomes in prostate VMAT treatments: a comparative study of machine learning models

• Machine learning models were developed to classify PSQA outcomes in prostate VMAT treatments. • Random Forest and Gradient Boosting achieved the highest accuracy and AUC values • The models effectively minimized false negatives, improving as the most predictive features. • Overlappingᵣectum, MCS and monitor Units were identified as the most predictive features. • The approach offers potential for real-time QA integration, reducing clinical workload and verification time. The present study aimed to construct and compare machine learning (ML) models for classifying patient-specific quality assurance (PSQA) outcomes in prostate volumetric modulated arc therapy (VMAT) treatment. A total of 1247 prostate VMAT plans were retrospectively analyzed and several metrics and anatomical information extracted from the RT-plans files and contours were used as predictive variables. The following machine learning (ML) models: Logistic Regression (LR), Decision Trees (DT), Random Forest (RF), Gradient Boosting (GB), AdaBoost (AB), k-Nearest Neighbors (KNN), Naïve Bayes (NB) and Neural Network (NN) were developed to classify the PSQA outcomes, and their performances compared with different metrics. The results demonstrated that Random Forest and Gradient Boosting achieved the highest classification accuracy with area under the curve (AUC) values of 0. 95 and 0. 96, respectively and Accuracy of 0. 91 for both classifiers. These models effectively balanced precision and accuracy while minimizing false negative and false positives rates which is critical for identifying potentially unsafe treatment plans. ML-based PSQA classification (Random Forest and Gradient Boosting) demonstrates strong potential for optimizing quality assurance in prostate VMAT treatments. By integrating predictive analytics into clinical workflows, radiation oncology departments can improve efficiency, reduce resource demands, and enhance patient safety, paving the way for more adaptive and automated QA protocols.