[Paper Review] A Review of Tomographic Reconstruction Techniques for Computed Tomography

Sanam Assili|arXiv (Cornell University)|Aug 28, 2018

Advanced X-ray and CT Imaging3 citations

TL;DR

This paper reviews analytical and iterative reconstruction techniques in computed tomography (CT), emphasizing filtered back-projection (FBP) and iterative methods like ART, SART, and MLEM. It highlights advancements in image quality and dose reduction through statistical and penalized likelihood approaches, demonstrating that iterative methods improve accuracy in low-dose and limited-data scenarios.

ABSTRACT

Medical imaging modalities have revolutionized health-care approaches by offering a better understanding of the human anatomy. Discovery of x-rays allowed the exploiting of the micro-scaled information of human anatomy. Computed tomography is one of the well-known imaging modalities using x-rays to create images of medical and non-medical objects. CT imaging has several variety of applications from medical diagnosis to industrial non-destructive testing. In this paper, we review computed tomography imaging modality and its applications in screening, diagnosis, and treatment and study some of the novel techniques used to reconstruct images in these machines.

Motivation & Objective

• To provide a comprehensive review of tomographic reconstruction techniques used in computed tomography (CT) for medical and industrial applications.
• To analyze the limitations of analytical methods like filtered back-projection (FBP) in handling measurement noise and artifacts.
• To explore iterative reconstruction techniques that improve image quality by modeling physical and statistical properties of the imaging process.
• To evaluate the role of advanced methods such as SART and MLEM in enabling low-dose CT imaging with reduced radiation exposure.
• To discuss emerging modalities like phase-contrast and dark-field imaging that extend CT capabilities beyond conventional attenuation-based imaging.

Proposed method

• Reviews the foundational principles of CT imaging, including x-ray generation via Bremsstrahlung, K-shell emission, and synchrotron sources.
• Describes the physical basis of x-ray attenuation through tissues using the photoelectric effect and Compton scattering.
• Explains analytical reconstruction using filtered back-projection (FBP), where a 1D filter is applied to projection data before back-projection into image space.
• Details iterative reconstruction methods such as Algebraic Reconstruction Technique (ART), which solves the linear system $AX = Y$ iteratively by minimizing residuals.
• Introduces Simultaneous Algebraic Reconstruction Technique (SART), which improves convergence and performance over ART by updating all pixels simultaneously.
• Examines statistical methods like Maximum Likelihood Expectation Maximization (MLEM) and Penalized Likelihood (PL), which model photon statistics and incorporate regularization to enhance image quality.

Experimental results

Research questions

• RQ1 How do analytical reconstruction techniques like FBP compare to iterative methods in handling noise and artifacts in CT imaging?
• RQ2 What are the key advantages of iterative reconstruction algorithms such as SART and MLEM in low-dose or limited-angle CT scenarios?
• RQ3 In what ways do statistical reconstruction methods improve image quality by modeling the physical measurement process?
• RQ4 How do emerging contrast mechanisms like phase-contrast and dark-field imaging expand the diagnostic potential of CT beyond conventional attenuation-based imaging?
• RQ5 What role do advances in computing power play in enabling the practical use of complex iterative reconstruction algorithms in clinical settings?

Key findings

• Filtered back-projection (FBP) remains the standard analytical method in commercial CT scanners due to its computational efficiency, but it is sensitive to noise and does not model measurement statistics.
• Iterative methods such as ART and SART provide improved image quality by iteratively refining image estimates based on comparison between simulated and measured projections.
• SART achieves good reconstruction quality in a single iteration and outperforms standard ART in convergence speed and stability, especially with limited projection data.
• Maximum Likelihood Expectation Maximization (MLEM) is effective in low-dose imaging by modeling photon-limited data and minimizing statistical uncertainty.
• Penalized likelihood (PL) methods enhance image reconstruction by incorporating regularization to reduce noise and preserve edges, balancing data fidelity and model complexity.
• Recent advances in computing power have enabled the practical implementation of iterative reconstruction, supporting dose reduction and improved diagnostic accuracy in clinical CT.