Four-Dimensional Cone Beam CT Imaging Using a Single Routine Scan via Deep Learning.

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Computational Imaging

Background:

• Four-dimensional cone-beam computed tomography (4D-CBCT) is crucial for motion management in radiotherapy.
• Acquiring high-quality 4D-CBCT from routine scans is challenging due to artifacts and motion.
• Existing methods struggle with severe streaking artifacts and accurate motion estimation.

Purpose of the Study:

• To develop a novel method for generating artifact-free 4D-CBCT images from a single routine scan.
• To improve the accuracy of deformable vector field (DVF) estimation for breathing phase reconstruction.
• To enhance the clinical feasibility of 4D-CBCT for upper abdominal cancer treatment.

Main Methods:

• Projections are sorted by diaphragm location to create phase-sorted data.
• A multiscale-discriminator generative adversarial network (MSD-GAN) is trained to reduce streaking artifacts.
• Deformable image registration estimates DVF, which is used in motion-compensated reconstruction for 4D-CBCT generation.

Main Results:

• The proposed MSD-GAN method significantly outperforms traditional iterative reconstruction and U-Net for 4D reconstruction quality.
• MSD-GAN demonstrates higher accuracy in image enhancement compared to U-Net.
• The method successfully generates high-quality 4D-CBCT images in both simulation and patient studies.

Conclusions:

• The novel MSD-GAN-based approach provides a practical and effective solution for 4D-CBCT imaging in upper abdominal cancer.
• This method significantly improves image quality and accuracy, aiding in precise radiotherapy delivery.
• The technique enables robust motion compensation, crucial for treating mobile tumors in the liver and pancreas.