Improving range estimation for carbon ion radiotherapy using artificial neural networks with Si/CdTe Compton camera.

Area of Science:

• Medical Physics
• Nuclear Instrumentation
• Computational Science

Background:

• Accurate monitoring of particle range is crucial for effective cancer radiotherapy.
• Traditional methods for range estimation in carbon-ion therapy have limitations in precision.
• Compton cameras offer potential for real-time monitoring but require sophisticated event selection.

Purpose of the Study:

• To improve carbon-ion range estimation accuracy using Compton imaging.
• To develop and validate an artificial neural network (ANN) for Compton event classification.
• To compare the ANN-based approach with traditional methods for range determination.

Main Methods:

• Monte Carlo simulations (Geant4) were used to model carbon-ion irradiation and gamma-ray detection.
• Silicon/cadmium telluride (Si/CdTe) Compton cameras detected emitted gamma-rays.
• An ANN was trained to distinguish true Compton events from background noise.
• List-mode maximum-likelihood expectation-maximization (LM-MLEM) reconstructed the photon distribution.
• Beam range was estimated by analyzing the emission profile along the beam axis.

Main Results:

• ANN-based event selection substantially increased the ratio of true Compton events.
• The proposed method achieved a mean position error below 3 mm across various phantom positions.
• This represents a significant improvement over the traditional method, which had a maximum error of 4.3 mm.

Conclusions:

• Integrating ANNs with Compton imaging enhances the precision of online range monitoring in carbon-ion therapy.
• The developed method shows promise for clinical applications in particle therapy.
• This approach offers improved accuracy and reliability for dose verification in radiation oncology.