Error detection for radiotherapy planning validation based on deep learning networks.

Area of Science:

• Medical Physics
• Radiotherapy
• Machine Learning

Background:

• Quality assurance (QA) for intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) requires plan validation.
• Current methods for analyzing positional dose distribution data lack sensitivity, hindering accurate identification of plan verification failures and complicating QA tasks.

Purpose of the Study:

• To develop a deep learning model for extracting 3D dose distribution maps.
• To create a predictive model for classifying errors in radiation therapy plans across various machine models, treatment techniques, and tumor sites.

Main Methods:

• Five categories of validation plans were created (normal, gantry error, collimator error, couch error, dose error).
• A Convolutional Neural Network (CNN) model was trained using 3D dose distribution data from 94 patients.
• Model performance was evaluated against the gamma pass rate (GPR) standard using different thresholds and tested on diverse accelerators.

Main Results:

• The CNN model achieved high performance metrics: accuracy (0.907), precision (0.925), recall (0.907), and F1 score (0.908).
• Comparable performance was observed on a different device (accuracy 0.900, precision 0.918, recall 0.900, F1 score 0.898).
• The CNN model demonstrated superior error prediction compared to the GPR method.

Conclusions:

• The CNN model offers superior predictive capability for radiation therapy plan validation compared to the GPR methodology, even across different devices.
• This deep learning approach enables faster and more efficient detection of plan validation failures, streamlining QA processes.
• The CNN model serves as a valuable tool to overcome limitations of the GPR method in radiation therapy QA.