Calculation of alpha particle single-event spectra using a neural network

Area of Science:

• Medical Physics
• Computational Biology
• Radiation Oncology

Background:

• Alpha-particle microdosimetry is crucial for radiation risk assessment.
• Predicting specific energy spectra is computationally intensive.
• Accurate spectral data is needed for advanced dosimetry calculations.

Purpose of the Study:

• To develop a neural network for predicting alpha-particle single-event specific energy spectra.
• To enhance the efficiency of microdosimetry calculations.
• To provide a tool for analyzing radiation interactions in biological targets.

Main Methods:

• A neural network with 4 inputs and 21 outputs was trained using Monte Carlo simulation data.
• Input parameters included source-target configuration, alpha particle energy, nuclei, and cell radii.
• The network utilized two hidden layers and mean square error loss function.

Main Results:

• The neural network achieved high accuracy, with root mean square errors (RMSE) for zmax below 1.57x10⁻².
• RMSE values for spectral outputs were consistently low across training, validation, and testing datasets.
• A correlation coefficient (R²) greater than 0.98 demonstrated strong agreement between predicted and actual spectral values.

Conclusions:

• The trained neural network accurately reproduces alpha-particle single-event spectra.
• This model is effective for a broad range of source-target geometries.
• The developed tool can significantly improve the speed and accuracy of microdosimetry calculations.