Enhancing patient-specific quality assurance for VMAT for breast cancer treatment: A machine learning approach for gamma passing rate (GPR) prediction

Abstract

Background

Modern radiation therapy for breast cancer has significantly advanced with the adoption of volumetric modulated arc therapy (VMAT), offering enhanced precision and improved treatment efficiency.

Purpose

To ensure the accuracy and precision of such complex treatments, a robust patient-specific quality assurance (PSQA) protocol is essential. This study investigates the potential of machine learning (ML) models to predict gamma passing rates (GPR), a key metric in PSQA.

Methods

A dataset comprising 863 VMAT plans was used to develop and compare seven ML models: Histogram-based gradient boosting regressor, random forest regressor, extra trees regressor, gradient boosting regressor, linear regression, AdaBoost regressor, and Multi-layer perceptron regressor. These models incorporated anatomical, dosimetric, and plan complexity features.

Results

Among the evaluated models, the extra trees regressor (ETR), random forest regressor (RFR), and gradient boosting regressor (GBR) demonstrated the best performance, achieving mean absolute errors (MAEs) of 0.51%, 0.52%, and 0.51%, and mean squared errors (MSEs) of 0.0051%, 0.0051%, and 0.0052%, respectively, on the validation dataset.

Conclusions

This study highlights the promise of ML-based approaches in streamlining PSQA processes, thereby supporting the quality assurance of breast cancer treatments using VMAT.