Convolution-based scatter correction using kernels combining measurements and Monte Carlo simulations

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Processing

Background:

• Scatter contamination is a significant challenge in Cone-Beam Computed Tomography (CBCT), degrading image quality and leading to inaccuracies.
• The scatter-to-primary ratio is particularly high in industrial applications, exacerbating image artifacts.

Purpose of the Study:

• To develop and validate an improved method for estimating scatter kernels in industrial CBCT.
• To compare a novel four-Gaussian scatter kernel model with a traditional two-Gaussian model.

Main Methods:

• Combined experimental measurements and Monte Carlo simulations to estimate scatter kernels.
• Utilized a continuously thickness-adapted kernels strategy with a four-Gaussian model.
• Compared the four-Gaussian model against an experimental two-Gaussian model.

Main Results:

• The four-Gaussian model demonstrated superiority in accounting for both object and detector scatter compared to the two-Gaussian model.
• Scatter kernels were parameterized with respect to object-to-detector distance.
• The proposed approach enables scatter kernel calculation across a wide range of magnifications.

Conclusions:

• A four-Gaussian scatter kernel model, validated through simulations and experiments, offers superior performance for industrial CBCT scatter correction.
• The developed parameterization simplifies scatter kernel estimation for diverse acquisition geometries.