Image Reconstruction from Sparse Low-Dose CT Data via Score Matching

W Cong, W Xia, G Wang - arXiv preprint arXiv:2306.08610, 2023 - arxiv.org
arXiv preprint arXiv:2306.08610, 2023arxiv.org
Computed tomography (CT) reconstructs sectional images from X-ray projections acquired
from multiple angles around an object. By measuring only a fraction of full projection data,
CT image reconstruction can reduce both radiation dose and scan time. However, with a
classic analytic algorithm, an image reconstructed from insufficient data contains severe
artifacts and loses fine structural details. To address this issue, various iterative and deep
learning reconstruction methods were developed to improve image quality. In this study, we …
Computed tomography (CT) reconstructs sectional images from X-ray projections acquired from multiple angles around an object. By measuring only a fraction of full projection data, CT image reconstruction can reduce both radiation dose and scan time. However, with a classic analytic algorithm, an image reconstructed from insufficient data contains severe artifacts and loses fine structural details. To address this issue, various iterative and deep learning reconstruction methods were developed to improve image quality. In this study, we present a deep learning-based image reconstruction method derived from maximum a posteriori (MAP) estimation. Our method utilizes a neural network to estimate the score function, which characterizes the probability density distribution of the image. The score function plays a crucial role in the Bayesian framework, contributing to the process of image reconstruction. The reconstruction algorithm theoretically guarantees the convergence of the iterative process. Our numerical results also show that this method produces decent sparse-view CT images.
arxiv.org