Fast prediction of patient-specific organ doses in brain CT scans using support vector regression algorithm

Area of Science:

• Medical Physics
• Radiology
• Computational Imaging

Background:

• Accurate patient-specific organ dose estimation is crucial for optimizing radiation protection in CT examinations.
• Traditional dose calculation methods can be computationally intensive, limiting real-time application.
• Radiomics offers a quantitative approach to extract features from medical images for predictive modeling.

Purpose of the Study:

• To develop and validate a support vector regression (SVR) model for predicting patient-specific head organ doses.
• To utilize radiomics features and graphics processing unit (GPU)-accelerated Monte Carlo (MC) simulations for dose calculation.
• To achieve rapid and accurate organ dose prediction in brain CT scans.

Main Methods:

• 237 patients undergoing brain CT scans were analyzed.
• Radiomics features were extracted from segmented head regions of interest (ROIs).
• GPU-accelerated MC simulations computed reference organ doses, and an SVR model was trained using these features and doses.

Main Results:

• The SVR model predicted head organ doses with a maximal difference of < 1 mGy compared to reference doses.
• Brain organ dose prediction achieved an absolute error of 1.3% and an R-squared value of 0.88.
• Predictions for eyes and lens showed RRMSE < 13%, MAPE of 4.5-5.5%, and R-squared > 0.7.

Conclusions:

• Patient-specific head organ doses from CT can be predicted rapidly (< 1 second) and accurately.
• The developed SVR model demonstrates high speed, accuracy, and robustness for clinical application.
• This approach enhances radiation safety by enabling efficient dose assessment.