Physics-Inspired Gaussian Kolmogorov-Arnold Networks for X-ray Scatter Correction in Cone-Beam CT

TL;DR

This work tackles scatter-induced bias and reduced tissue contrast in cone-beam CT by introducing a physics-informed deep learning approach that models projection-domain scatter with Gaussian radial basis functions embedded in Kolmogorov–Arnold Networks. The method maps scatter-contaminated projections to a scatter estimate and performs denoising before reconstructing scatter-free images, with a Kn-based Gaussian RBF layer enabling efficient nonlinear feature extraction that respects the Klein-Nishina symmetry. Evaluations on synthetic data show superior scatter estimation and reconstructed image quality compared with SKS, DSE-Net, UNet-ViT, and Pix2pix GAN, with notable gains in PSNR, SSIM, and HU accuracy. The results suggest a practical, model-driven hybrid approach capable of closer-to-ground-truth reconstructions and potential real-time CBCT scatter correction in clinical workflows.

Abstract

Cone-beam CT (CBCT) employs a flat-panel detector to achieve three-dimensional imaging with high spatial resolution. However, CBCT is susceptible to scatter during data acquisition, which introduces CT value bias and reduced tissue contrast in the reconstructed images, ultimately degrading diagnostic accuracy. To address this issue, we propose a deep learning-based scatter artifact correction method inspired by physical prior knowledge. Leveraging the fact that the observed point scatter probability density distribution exhibits rotational symmetry in the projection domain. The method uses Gaussian Radial Basis Functions (RBF) to model the point scatter function and embeds it into the Kolmogorov-Arnold Networks (KAN) layer, which provides efficient nonlinear mapping capabilities for learning high-dimensional scatter features. By incorporating the physical characteristics of the scattered photon distribution together with the complex function mapping capacity of KAN, the model improves its ability to accurately represent scatter. The effectiveness of the method is validated through both synthetic and real-scan experiments. Experimental results show that the model can effectively correct the scatter artifacts in the reconstructed images and is superior to the current methods in terms of quantitative metrics.

Figures (6)

• Figure 1: Schematic diagram of two-dimensional scatter distribution characteristics in a three-dimensional CBCT system.
• Figure 2: The overall architecture of the proposed method.The dotted blue part is the encoding part, and the dotted pink part is the decoding part.
• Figure 3: The overall architecture of Gaussian KAN layer.
• Figure 4: Synthetic experiment results on the estimated scatter signal using different methods. From left to right are Reference, SKS, DSE-Net, UNet-ViT, Pix2pix GAN, and our method. The first row displays the projection domain scatter images, while the second row shows the difference images relative to the reference image. Display windows: $[300, 1400]$ photons.
• Figure 5: Synthetic experiment reconstructed results on scatter correction using compared methods. The first row is reconstructed images, while the second row shows the difference images relative to the reference image. The display window of uncorrected error image set [-100, 800] HU. The others display windows are defined as [-80, 80] HU.