Deep learning estimation of proton stopping power with photon-counting computed tomography: a virtual study

Area of Science:

• Medical Physics
• Radiology
• Artificial Intelligence

Background:

• Proton therapy offers precise tumor targeting with reduced normal tissue damage due to proton Bragg peaks.
• Accurate proton stopping power ratio (SPR) estimation from CT images is crucial for aligning the high-dose region with the tumor.
• Photon-counting CT (PCCT) offers superior quantitative imaging and resolution compared to conventional CT.

Purpose of the Study:

• To evaluate the potential of photon-counting CT (PCCT) for enhancing SPR estimation.
• To develop and train a deep neural network for transforming PCCT images into SPR maps.
• To assess the accuracy of AI-driven SPR estimation compared to traditional methods.

Main Methods:

• Simulated PCCT head images and ground truth SPR maps were generated using the XCAT phantom and CatSim software.
• A U-Net deep neural network was trained using simulated PCCT images as input and SPR maps as labels.
• The network was trained with specific parameters: 260 mA tube current, 120 kV tube voltage, and 4000 view angles.

Main Results:

• The deep neural network achieved an average root mean square error (RMSE) of 0.26%–0.41% for SPR prediction.
• This represents a significant improvement over physical modeling methods for single-energy CT (0.40%–1.30% RMSE) and dual-energy CT (0.41%–3.00% RMSE).

Conclusions:

• Combining PCCT with deep learning shows significant promise for accurate SPR estimation.
• This approach has the potential to reduce beam range uncertainties in proton therapy.
• Further development could lead to more precise and effective proton radiation treatments.