Author: Hua-Chieh Shao, Guoping Xu, You Zhang, Tingliang Zhuang 👨🔬
Affiliation: 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 🌍
Purpose:
Advancements in onboard X-ray hardware allow high-quality CBCT imaging with a short scan time (~6s for Varian HyperSight), enabling CBCT-based dose calculation and treatment planning. However, for respiratory motion, more than one breathing cycle still occurs within the scan, leading to motion blurriness and target/normal-tissue definition uncertainties. We propose a dynamic CBCT reconstruction framework for fast imaging systems to resolve intra-scan motion, by combining optical surface and X-ray imaging (Surf-X-recon).
Methods:
Surf-X-recon reconstructs CBCTs for each X-ray projection to capture intra-scan motion, combining optical surface and X-ray imaging for enhanced accuracy and robustness. Surf-X-recon reconstructs a dynamic sequence of CBCTs through a joint image reconstruction and deformable registration approach that simultaneously solves a reference CBCT and a patient-specific motion model. Dynamic CBCTs are then generated by deforming the reference CBCT with time-varying motion fields derived from the motion model. Implicit neural representation is used to model/reconstruct the reference CBCT, while the motion model is solved on-the-fly as learnable B-splines in a data-driven manner. To stabilize the solution of dynamic CBCTs and the motion model, body surface maps, captured concurrently with X-ray via the optical imaging system, are integrated into the fidelity loss function as complementary information to constrain and condition the reconstruction.
Results:
Evaluated via a simulation study from real-patient 4D-CT scans, surf-X-recon can reconstruct accurate motion-resolved dynamic CBCTs and motion models. The incorporation of surface imaging improves image quality and motion model accuracy by removing unreasonable deformations due to motion/anatomy ambiguities and under-sampling. After incorporating surface imaging, the diaphragm apex tracking error (Mean±S.D.) reduced from 2.3±1.4mm (without surface imaging) to 1.9±0.9mm, and the body surface error reduced from 1.3±0.6mm to 0.7±0.2mm.
Conclusion:
Surf-X-recon can reconstruct a dynamic sequence of CBCTs for a 6s HyperSight scan with joint X-ray and surface imaging, providing essential motion information for image-guided radiotherapy.