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

PURPOSE: This study aims to enhance carbon-ion range estimation using a Compton camera, improved through artificial neural network (ANN)-based event classification.

METHODS

A Monte Carlo simulation was developed in Geant4. Carbon ions irradiated to a PMMA phantom and 511 keV gamma-rays emitted were detected by silicon/cadmium telluride (Si/CdTe) Compton cameras. An ANN was trained to identify true Compton events. Following event selection, the annihilation photon distribution was reconstructed using the list-mode maximum-likelihood expectation-maximization algorithm. The beam range was estimated by identifying the peak position of the emission profile along the beam axis. This method was compared to traditional approach based on geometric constraints and energy thresholds.

RESULTS

ANN-based event selection significantly improved the true Compton event ratio and the accuracy of range estimation. Across a phantom position range from -15 mm to 30 mm, the proposed method achieved a phantom position mean error of less than 3 mm, outperforming the traditional method, which showed a maximum error of 4.3 mm.

CONCLUSIONS

The integration of ANN-based event classification with Compton imaging improves the precision of online range monitoring for carbon-ion therapy. This approach demonstrates strong potential for clinical application, offering enhanced accuracy and reliability in particle therapy dose verification.