Predicting patient-specific organ doses from thoracic CT examinations using support vector regression algorithm.

Area of Science:

• Medical Physics
• Radiology
• Computational Imaging

Background:

• Accurate patient-specific organ dose estimation in CT examinations is crucial for radiation protection.
• Traditional methods like Monte Carlo simulations are computationally intensive and time-consuming.

Purpose of the Study:

• To develop a fast, accurate, and robust prediction method for patient-specific organ doses from CT examinations.
• To minimize computational resources required for dose prediction.

Main Methods:

• Auto-segmentation of thoracic organs from CT images to define regions of interest (ROIs).
• Extraction of radiomics features from ROIs using the Pyradiomics package.
• Training a support vector regression (SVR) model with radiomics features and Monte Carlo simulated organ doses.

Main Results:

• The SVR model achieved high accuracy with R-squared values from 0.75 to 0.89 on test sets.
• Low root mean squared errors (RMSE) between 2-2.8 mGy (test sets) and mean absolute percentage errors (MAPE) of 0.1-0.18 (test sets).
• The method demonstrated robustness through cross-validation and achieved prediction in under one second per patient.

Conclusions:

• A combination of SVR and thoracic radiomics features enables accurate, fast, and robust prediction of patient-specific organ doses.
• The developed method significantly reduces computational demands, making it practical for clinical use.
• This approach offers a promising tool for optimizing radiation safety in CT imaging.