CBCT correction using a cycle-consistent generative adversarial network and unpaired training to enable photon and...

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Artificial Intelligence in Medicine

Background:

• Image-guided adaptive radiotherapy requires accurate patient imaging.
• Cone-beam CT (CBCT) is used for pre-treatment imaging but suffers from artifacts, limiting its use for dose calculation.
• Deep learning offers potential for CBCT image correction.

Purpose of the Study:

• To investigate the feasibility of using a cycle-consistent generative adversarial network (cycleGAN) for correcting prostate CBCT images.
• To assess the accuracy of CBCT images corrected by cycleGAN for photon and proton therapy dose calculations.
• To evaluate the time efficiency of the cycleGAN correction method.

Main Methods:

• A cycleGAN was trained using unpaired original CBCT (CBCTorg) and planning CT equivalent images (CBCTcycleGAN) from 33 patients.
• HU accuracy was compared to a validated correction technique (CBCTcor).
• Dosimetric accuracy was evaluated for volumetric-modulated arc photon therapy (VMAT) and opposing single-field uniform dose (OSFUD) proton plans, and proton range accuracy was assessed.

Main Results:

• The mean HU error decreased from 24 HU (CBCTorg) to -6 HU (CBCTcycleGAN) compared to CBCTcor.
• High dose calculation accuracy was achieved for VMAT (100%/89% pass-rate for 2%/1% dose difference).
• For proton OSFUD plans, 80% pass-rate for 2% dose difference and 96% for (2%, 2mm) gamma criterion were achieved. 93% of SFUD profiles had range agreement within 3mm.
• CBCT correction time reduced from 6-10 min to 10 seconds.

Conclusions:

• CycleGAN is feasible for prostate CBCT correction, offering significant speed-up for adaptive radiotherapy.
• High dose calculation accuracy was demonstrated for VMAT plans.
• Further improvements may be needed for proton therapy applications.
• The unpaired training approach avoids reliance on anatomically consistent data or deformable image registration.