Time-resolved dynamic CBCT reconstruction using prior-model-free spatiotemporal Gaussian representation (PMF-STGR)

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Computational Imaging

Background:

• Dynamic cone-beam CT (CBCT) imaging is crucial for characterizing intra-scan motion, enabling precise patient setup and motion-adapted radiotherapy.
• Current methods for dynamic CBCT reconstruction face challenges in speed and accuracy, limiting their clinical applicability.

Purpose of the Study:

• To develop and evaluate a novel Gaussian representation-based framework (PMF-STGR) for fast and accurate time-resolved CBCT reconstruction.
• To model intra-scan patient anatomy and motion using 3D Gaussians for improved dynamic CBCT accuracy.

Main Methods:

• Developed PMF-STGR using a dense set of 3D Gaussians for reference CBCT and another set for motion-basis-components (MBCs).
• Employed a CNN-based motion encoder to determine temporal coefficients for MBCs, generating deformation vector fields.
• Reconstructed dynamic CBCTs in a 'one-shot' training manner from standard 3D CBCT scans without prior models.

Main Results:

• PMF-STGR demonstrated superior performance compared to an implicit neural representation (INR)-based approach (PMF-STINR) in XCAT phantom simulations and patient scans.
• Achieved approximately 50% reduction in reconstruction time with improved image quality (lower RE, higher SSIM) and comparable/better motion accuracy.
• Mean RE, SSIM, and COME for PMF-STGR were 0.128(0.009), 0.990(0.002), and 0.71 mm(0.40 mm), respectively, outperforming PMF-STINR.

Conclusions:

• PMF-STGR offers enhanced efficiency and accuracy for dynamic CBCT reconstruction.
• The framework's performance suggests significant potential for clinical translation in radiotherapy applications.
• Improved dynamic CBCT imaging facilitates better motion characterization and adaptive treatment strategies.