Author: Hua-Chieh Shao, Shanshan Tang, Jing Wang, Kai Wang, You Zhang 👨🔬
Affiliation: Department of Radiation Oncology, UT Southwestern Medical Center, Medical Artificial Intelligence and Automation (MAIA) Lab, Department of Radiation Oncology, UT Southwestern Medical Center, Medical Artificial Intelligence and Automation (MAIA) Laboratory, Department of Radiation Oncology, UT Southwestern Medical Center, Department of Radiation Oncology, University of Maryland Medical Center 🌍
Purpose: Artifacts caused by gas bubble movement in the gastrointestinal tract can severely degrade the image quality of on-board abdominal cone-beam computed tomography (CBCT), impacting its utility in target locating, patient setup, and adaptive planning. This study introduces a novel hybrid framework combining Prior Model-Free Spatiotemporal Implicit Neural Representation (PMF-STINR) and Simultaneous Motion Estimation and Image Reconstruction (SMEIR) techniques to address gas bubble motion artifacts. By leveraging the strengths of both methods, we propose STINR-SMEIR which aims to achieve high-quality abdominal CBCT imaging for improving radiotherapy workflow.
Methods: In the proposed STINR-SMEIR, deep learning based PMF-STINR method is first employed to reconstruct low-resolution dynamic CBCT from singular x-ray projections and estimate corresponding deformation vector fields (DVFs) to baseline CBCT, which capture spatiotemporal motion patterns during CBCT acquisition. These DVFs are used for projection group partition and serve as initialization for SMEIR, which is an iterative reconstruction method. SMEIR iteratively refines the DVFs between each projection group and reconstructs high-resolution artifact-reduced CBCT images. The framework is evaluated with simulated projections of a clinical abdominal 4D-CT. Image reconstruction fidelity analysis and qualitive image assessment are performed to compare CBCTs reconstructed with different methods to original 4D CT.
Results: The STINR-SMEIR method demonstrated improved ability to achieve higher image resolution and HU uniformity while reducing the artifacts around gas bubble region. Image similarity quantified by SSIM (0.9411) and PNSR (33.56) showed STINR-SMEIR reconstructed CBCT has the superior similarity to the original image.
Conclusion: The proposed hybrid framework provides an effective solution for correcting gas bubble motion artifacts in on-board abdominal CBCT. It shows the potential to handle the irregular and aperiodic gas bubble motion while maintaining higher image resolution, which is typically challenging for traditional methods.