Dose image prediction for range and width verifications from carbon ion-induced secondary electron bremsstrahlung...

Area of Science:

• Medical Physics
• Radiotherapy
• Image Analysis

Background:

• Secondary electron bremsstrahlung (SEB) x-ray imaging aids particle-ion beam range estimation.
• Limitations include poor correlation with dose images and low spatial resolution.
• Accurate beam range and width determination is crucial for effective radiotherapy.

Purpose of the Study:

• To develop a deep learning (DL) model for predicting dose images from SEB x-ray images.
• To overcome limitations of SEB x-ray imaging for accurate carbon ion beam range and width estimation.
• To validate the DL model's performance on both simulated and measured data.

Main Methods:

• Generated 10,000 simulated SEB x-ray and dose image pairs for DL training.
• Employed a DL neural network with two U-Nets for image conversion and super-resolution.
• Evaluated dose image prediction accuracy using Mean Squared Error (MSE) and Structural Similarity Index (SSIM).

Main Results:

• Simulated data: MSE of 2.5 × 10⁻⁵, SSIM of 0.997; range/width error within 1 mm FWHM.
• Measured data: MSE ≤ 5.5 × 10⁻³, SSIM ≥ 0.980; range/width error of 2 mm and 5 mm FWHM.
• DL approach demonstrated effective dose image prediction from SEB x-ray images.

Conclusions:

• The deep learning approach successfully predicts dose images from SEB x-ray images.
• This method enhances the accuracy of carbon ion beam range and width estimation.
• The DL model shows promise for improving radiotherapy planning and delivery.