Scatter correction for cone-beam CT via scatter kernel superposition-inspired convolutional neural network

Abstract

{Objective: }X-ray scatter leads to signal bias and degrades the image quality in CT imaging. Conventional real-time scatter estimation and correction methods include the scatter kernel superposition (SKS) methods, which approximate X-ray scatter field as a convolution of the scatter sources and scatter propagation kernels to reflect the spatial spreading of scatter X-ray photons. SKS methods are fast to implement but generally suffer from low accuracy due to the difficulties in determining the scatter kernels. {Approach: }To address such a problem, this work describes a new scatter estimation and correction method by combining the concept of SKS methods and convolutional neural network. Unlike conventional SKS methods which estimate the scatter amplitude and the scatter kernel based on the value of an individual pixel, the proposed method generates the scatter amplitude maps and the scatter width maps from projection images through a neural network, from which the final estimated scatter field is calculated based on a convolution process. {Main Results: }By incorporating physics in the network design, the proposed method requires fewer trainable parameters compared with another deep learning-based method (Deep Scatter Estimation). Both numerical simulations and physical experiments demonstrate that the proposed SKS-inspired convolutional neural network outperforms the conventional SKS method and other deep learning-based methods in both qualitative and quantitative aspects. {Significance: } The proposed method can effectively correct the scatter-related artifacts with a SKS-inspired convolutional neural network design.

Asymmetric scatter kernel superposition-inspired deep learning approach to estimate scatter in breast tomosynthesis

“…The structure of the asymmetric kernel-based network is illustrated in Figure 1, presenting a modified version of Zhuo's previous work [8]. This network features two input branches: the upper branch, which integrates the projection angle and Euclidean distance map to account for changing thickness and shape, and the lower branch, where the concatenation of log-transformed projection and its natural logarithm is utilized to estimate the scatter amplitude map.…”

“…Software-based scatter estimation/correction approaches include the Monte Carlo-based (MC) method [3], the fast adaptive scatter kernel superposition (fASKS) method [4,5], and the deep learning approaches [6][7][8]. While MC simulation is regarded as the most accurate method to estimate scatter, it is computationally intensive and too slow for clinical use.…”

“…However, existing DL methods follow an end-to-end training strategy that could overlook the underlying physics of scatter formation, and thus might cause secondary artifacts. To leverage the interpretability of kernel-based scatter correction methods, Zhuo et al [8] propose a symmetric scatter kernel-inspired deep learning approach for cone-beam CT. In contrast to most existing studies where deep learning is utilized to directly obtain scatter field as the network output, they obtain the scatter amplitude map and scatter kernel width map through the network.…”

Task-based transferable deep-learning scatter correction in cone beam computed tomography: a simulation study

“…9 Particularly, in the field of medical imaging, previous works have demonstrated the potential of CNN to perform x-ray scatter corrections. [9][10][11][12][13][14][15][16][17][18] However, in real clinical practice, there are still challenges to effectively use CNN to compensate for x-ray scatter. First, it is often the case that deep learning models benefit from large volumes of data, which are not always trivial to obtain or curate, especially for clinical applications, where patient safety and consent is a priority.…”