A hybrid method of correcting CBCT for proton range estimation with deep learning and deformable image registration

Area of Science:

• Medical Physics
• Radiotherapy
• Image Processing

Background:

• Accurate synthetic CT (sCT) generation is crucial for proton therapy planning and verification.
• Bowel gas pockets in cone-beam CT (CBCT) pose challenges for sCT accuracy and proton range estimation.
• Existing methods struggle with the complex interplay of high and low spatial frequencies in CBCT data.

Purpose of the Study:

• To develop and validate a novel hybrid method for generating sCT from abdominal/pelvic CBCT, specifically addressing challenges posed by bowel gas.
• To improve the accuracy of proton range estimation in regions with gas pockets.
• To enhance the utility of CBCT for adaptive replanning in proton therapy.

Main Methods:

• A hybrid approach combining unsupervised deep learning (CycleGAN) and deformable image registration (DIR) was developed.
• CycleGAN generated the geometry-weighted (high-frequency) component, while DIR handled the intensity-weighted (low-frequency) component of the sCT.
• Iterative feedback loops refined the sCT, particularly in bowel gas regions, by adjusting DIR and incorporating deformed planning CT (pCT) information.

Main Results:

• The hybrid sCT demonstrated significantly higher accuracy in CT numbers (lower mean absolute error) compared to deformed pCT and CycleGAN-only sCT.
• Internal gas regions showed significantly higher Dice similarity with the hybrid method.
• Proton range estimation errors were significantly reduced, and gamma passing rates showed a dosimetric advantage with the hybrid method.

Conclusions:

• The hybrid method significantly improves sCT accuracy from CBCT, outperforming individual deep learning and registration techniques.
• This approach shows promise for accurate proton range verification and adaptive replanning in abdominal/pelvic proton therapy, even in the presence of bowel gas.
• The study highlights the potential of hybrid image processing techniques for advancing radiotherapy applications.