Technical Note: Machine learning approaches for range and dose verification in proton therapy using proton-induced...

Area of Science:

• Medical Physics
• Radiotherapy
• Machine Learning

Background:

• Proton therapy offers precise radiation delivery but requires robust quality assurance.
• Online verification of proton range and dose is crucial for patient safety.
• Positron-emitting isotopes produced during proton therapy can be measured for verification.

Purpose of the Study:

• To develop and evaluate a machine learning framework for online proton range and dose verification.
• To establish the relationship between positron emitter activity distributions and dose distributions.
• To assess the performance of neural network models under varying noise conditions.

Main Methods:

• Simulations using GATE and Geant4 with a CT-based patient phantom.
• Development of a feedforward neural network for range verification (classification).
• Development of a recurrent neural network (RNN) for dose distribution prediction (regression).
• Evaluation using mean squared error (MSE) and mean absolute error (MAE) at different signal-to-noise ratios (SNRs).

Main Results:

• The neural network framework demonstrated feasibility for proton range and dose verification.
• Feedforward NN achieved near 100% classification accuracy for range verification.
• RNN accurately predicted 1D dose distributions across energies and positions.
• With SNR > 4, predictions achieved MAE of ~0.60 mm and MSE of ~0.066.
• The models showed good generalization capabilities.

Conclusions:

• Recurrent neural networks effectively model the relationship between positron emitter and dose distributions.
• The proposed machine learning framework, particularly RNNs, shows promise for online verification in proton therapy.
• This approach can enhance quality assurance by enabling accurate real-time range and dose verification.