Error detection using a convolutional neural network with dose difference maps in patient-specific quality assurance...

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Artificial Intelligence in Healthcare

Background:

• Patient-specific quality assurance (QA) is crucial for volumetric modulated radiation therapy (VMAT).
• Detecting multi-leaf collimator (MLC) positional errors ensures treatment accuracy.
• Current methods like gamma analysis have limitations in error detection.

Purpose of the Study:

• To evaluate the efficacy of convolutional neural networks (CNNs) using dose difference maps for detecting MLC positional errors.
• To assess the performance of CNNs in patient-specific QA for VMAT.
• To compare the CNN approach with traditional gamma analysis.

Main Methods:

• Dose difference maps were generated from measured and calculated dose distributions for VMAT plans.
• A three-dimensional CNN model was trained and tested using simulated error-free and erroneous (systematic and random MLC errors) datasets.
• The CNN model was validated using five-fold cross-validation.

Main Results:

• The CNN model achieved an overall accuracy of 0.944 in classifying plans as error-free, systematic error, or random error.
• High sensitivity (0.889-1.000) and specificity (0.944-0.986) were observed for all classifications.
• The CNN approach demonstrated superior performance compared to gamma analysis.

Conclusions:

• Dose difference maps combined with CNNs offer an effective method for detecting MLC errors in VMAT patient-specific QA.
• This AI-driven approach enhances the reliability and accuracy of radiation therapy quality assurance.
• The study highlights the potential of CNNs to improve safety and precision in VMAT delivery.