Deep learning-based dose prediction for low-energy electron beam superficial radiotherapy.

BACKGROUND

Accurate surface dose calculation is crucial in superficial low-energy electron beam radiotherapy owing to shallow treatment depths and the risk of skin toxicity. Traditional Monte Carlo (MC) simulations are precise but computationally expensive and time-consuming.

METHODS

This study combined MC simulations with deep learning to improve both accuracy and speed. DOSXYZnrc was used to simulate low-energy electron beams for six body sites, generating computed tomography phantoms and corresponding dose distributions. A cascaded 3D U-Net (C3D) model was trained on these datasets to predict dose distributions rapidly.

RESULTS

The C3D model demonstrated significant improvements over traditional 3D U-Net models, achieving a minimum Gamma pass rate of 92.09% and a minimum dose difference pass rate of 93.58%. The model completed dose predictions in just 0.42 seconds, making predictions approximately 140,000 times faster than MC simulations. In the evaluation of dose distributions across six anatomical regions, C3D consistently outperformed other deep learning models (3D U-Net, Deep Convolutional Neural Network, and HD U-Net) in both accuracy and robustness.

CONCLUSION

The integration of deep learning with MC simulations significantly enhances the efficiency of surface dose calculations in superficial electron beam radiotherapy. The C3D model provides rapid and accurate dose predictions, facilitating efficient treatment planning while maintaining high accuracy.