Research on the error pattern recognition of dosimetric quality assurance by Bayesian optimization

Area of Science:

• Medical Physics
• Radiotherapy
• Quality Assurance

Background:

• Improving dose delivery accuracy in radiotherapy is crucial for clinical benefits.
• Efficient detection of multiple error types in dosimetric quality assurance (DQA) data remains a challenge.
• This study addresses the need for advanced methods to analyze complex DQA data.

Purpose of the Study:

• To develop and validate an algorithm for quantitatively analyzing multiple errors in DQA data.
• To leverage Bayesian optimization (BO) and statistical methods for enhanced error detection.
• To improve the accuracy of dose delivery in radiotherapy through effective error identification.

Main Methods:

• Utilized Bayesian optimization (BO) with a Gaussian process (GP) model to adjust error matrices and minimize DQA failure rates.
• Analyzed 79 treatment plans from an Infinity linear accelerator (LINAC), including errors from MLC, jaws, and collimator rotation.
• Evaluated the algorithm using simulated data with known error magnitudes and real-world clinical DQA data.

Main Results:

• The developed algorithm accurately detected simulated systematic errors, with detected matrices closely matching introduced values.
• Correction of inherent systematic errors in clinical data led to significant reductions in DQA failure rates.
• Failure rates decreased from 6.06% to 1.78% in the training set and 4.15% to 2.02% in the testing set.

Conclusions:

• The error pattern recognition algorithm effectively detects and quantifies multiple error types in DQA data.
• This method enhances radiotherapy plan implementation accuracy and can identify systematic deviations in clinical DQA.
• The algorithm provides valuable labeled datasets for deep learning applications in radiotherapy quality assurance.