Machine-learning-based integration of temporal and spectral prompt gamma-ray information for proton range...

Area of Science:

• Medical Physics
• Radiation Oncology
• Nuclear Instrumentation

Background:

• Prompt gamma-ray timing (PGT) and spectroscopy (PGS) are key non-invasive methods for monitoring proton radiotherapy.
• Integrating PGT and PGS may enhance proton range verification by combining temporal and spectral data.
• Machine learning is explored to improve accuracy in proton range verification.

Purpose of the Study:

• To evaluate the effectiveness of integrating prompt gamma-ray timing and spectroscopy data for enhanced proton range verification.
• To assess the contribution of spectral information compared to temporal information in machine learning models for range shift prediction.
• To determine if temporal information alone is sufficient for accurate proton range verification.

Main Methods:

• A homogeneous phantom was irradiated with proton beams (162 and 225 MeV).
• Simulated anatomical variations (air cavities) were introduced.
• Prompt gamma rays were measured using a PGT detector, extracting 2D time-energy spectra.
• Machine learning models were trained using different feature sets (energy-only, time-only, combined, image) to predict range shifts, with performance assessed by RMSE.

Main Results:

• Time-only and combined time-energy features achieved good performance (RMSE 3-4 mm).
• Energy-only and image features showed poorer performance (RMSE > 5 mm).
• Integrating energy-only features did not enhance prediction accuracy over time-only features.

Conclusions:

• Spectral information did not provide additional value for proton beam range shift determination in this setup.
• Temporal information from prompt gamma rays alone is sufficient for accurate proton range verification.
• This study validates the use of PGT for reliable range verification in proton therapy.