Machine learning prediction of effective radiation doses in various computed tomography applications: a virtual human...

Area of Science:

• Medical Physics
• Radiological Sciences
• Computational Imaging

Background:

• Computed tomography (CT) scans are widely used diagnostic tools.
• Accurate estimation of radiation dose is crucial for patient safety in CT imaging.
• Variations in patient weight and CT protocols significantly influence radiation exposure.

Purpose of the Study:

• To employ machine learning (ML) algorithms for precise forecasting of radiation doses in CT scans.
• To evaluate the performance of different ML models in predicting effective doses across diverse CT protocols and phantom characteristics.
• To identify the most effective ML approach for CT radiation dose estimation.

Main Methods:

• Utilized cloud-based software to compute effective doses for various CT protocols.
• Employed eight full-body computational phantoms representing a spectrum of adult body weights.
• Developed a dataset integrating head, neck, and chest-abdomen-pelvis CT scan parameters.
• Applied Linear Regression (LR), Random Forest (RF), and Support Vector Regression (SVR) ML algorithms.
• Evaluated model performance using mean absolute error, mean squared error, and accuracy metrics.

Main Results:

• Female phantoms absorbed 7.8% higher radiation doses compared to male phantoms.
• Radiation dose increased progressively from overweight to normal weight phantoms (11% increments).
• Linear Regression (LR) achieved a 0% error rate and 100% accuracy in predicting CT radiation doses.
• Significant dose variations were observed based on phantom weight and CT scan protocols.

Conclusions:

• Linear Regression (LR) emerged as the superior ML algorithm for estimating CT-induced radiation doses.
• ML offers a promising avenue for accurate and reliable CT radiation dose prediction.
• Findings highlight the importance of considering patient-specific factors (weight) and protocol variations in dose assessment.