Development of a patient-specific cone-beam computed tomography dose optimization model using machine learning in...

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Machine Learning in Healthcare

Background:

• Cone-beam computed tomography (CBCT) is crucial for soft tissue and bone visualization in radiation therapy.
• Optimizing CBCT dose is essential to reduce patient radiation exposure without compromising image quality for treatment planning and delivery.

Purpose of the Study:

• To develop and validate a machine learning model for predicting patient-specific CBCT doses.
• To minimize radiation dose while maintaining clinically acceptable soft tissue image quality in prostate radiation therapy.

Main Methods:

• Phantom studies assessed the relationship between radiation dose and image quality metrics (standard deviation, contrast-to-noise ratio).
• Clinical CBCT and planning CT data were used to train and test machine learning models.
• Support vector regression was selected for its high accuracy in dose prediction.

Main Results:

• A minimum clinical dose level of 40% was identified, showing minimal CNR decrease in phantom studies.
• Image quality metrics degraded with decreasing dose, consistent with phantom findings.
• The machine learning model achieved an R² of 0.833 for predicting CBCT doses.

Conclusions:

• Patient-specific CBCT imaging protocols can significantly reduce radiation dose.
• The developed machine learning model enables dose reduction while preserving clinically acceptable image quality for soft tissue registration.
• This approach supports image-guided radiotherapy with reduced cumulative radiation exposure.