Spatiotemporal Gaussian Optimization for 4D Cone Beam CT Reconstruction from Sparse Projections.

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Image Reconstruction

Background:

• Four-dimensional cone-beam computed tomography (4D-CBCT) is essential for image-guided radiotherapy (IGRT) to track tumor motion during breathing.
• Current 4D-CBCT methods require extensive projection data, leading to long scan times and increased patient radiation dose.
• Reconstructing high-quality 4D-CBCT from limited projections is challenging due to sparse sampling and resultant artifacts.

Purpose of the Study:

• To develop a novel framework for reconstructing high-quality 4D-CBCT images from sparse projection data within a 1-minute acquisition.
• To effectively reduce streak artifacts while preserving dynamic motion and fine spatial details in 4D-CBCT.
• To enable faster and lower-dose 4D-CBCT imaging for improved IGRT.

Main Methods:

• A spatiotemporal Gaussian representation framework was introduced, where each Gaussian is defined by position, covariance, rotation, and density.
• 2D X-ray projections were rendered from the Gaussian point cloud using X-ray rasterization.
• A Gaussian deformation network was jointly optimized to model dynamic CBCT scenes, followed by voxelization to reconstruct 4D-CBCT images.

Main Results:

• The proposed method successfully reconstructed high-quality 4D-CBCT images from sparse projections.
• The framework achieved a balance between streak artifact reduction, dynamic motion preservation, and fine detail restoration.
• The reconstructed images demonstrated superior quality compared to standard methods under sparse sampling conditions.

Conclusions:

• Spatiotemporal Gaussian representation offers a promising solution for efficient and high-quality 4D-CBCT reconstruction from sparse data.
• This technique can significantly improve the feasibility of 4D-CBCT in clinical IGRT by reducing scan time and dose.
• The developed framework holds potential for advancing adaptive radiotherapy and motion management.