Potential of Metal Artifact Reduction (MAR) and Deep Learning-based Reconstruction (DLR) algorithms integration in CT Metal Artifact Correction: a review

Mohammed Musta, Njiti and Osman, Noor Diyana and Mansor, Mohd Syahir and Rabaiee, Nor Ain and Abdul Aziz, Mohd Zahri (2024) Potential of Metal Artifact Reduction (MAR) and Deep Learning-based Reconstruction (DLR) algorithms integration in CT Metal Artifact Correction: a review. Radiation Physics and Chemistry, 218. pp. 1-15. ISSN 0969-806X E-ISSN 1879-0895

	PDF - Published Version Restricted to Repository staff only Download (2MB) | Request a copy
Preview	PDF - Supplemental Material Download (261kB) | Preview

Abstract

Computed Tomography (CT) is essential for precise medical diagnostics, yet metal implants often induce disruptive image artifacts. Metal Artifact Reduction (MAR) algorithms have emerged to enhance CT image quality by mitigating these artifacts. This review emphasizes the significance of quantifying MAR algorithms, details common quantification metrics, and presents findings from diverse CT scanner studies. MAR techniques effectively reduce metal artifacts and enhance CT imaging. Metrics like noise levels, Contrast-to-Noise ratio (CNR), CT number accuracy, and Metal Artifact Index (MAI) quantify their efficacy. Varied CT scanner experiments with diverse metal implants display improved CT number accuracy, noise reduction, and artifact management through MAR algorithms. However, secondary artifacts and altered metal size accuracy are potential drawbacks that need attention. Deep Learning-based Reconstruction (DLR) is an expanding approach using Artificial Intelligence (AI) for CT image reconstruction. DLR generates low-dose CT images with high spatial resolution. Recent clinical deployments highlight DLR’s potential in generating low-noise, texture-rich images, and superior artifact reduction. Moreover, DLR techniques exhibit promise in addressing beam hardening artifacts. While MAR algorithms have revolutionized CT imaging, DLR techniques are emerging as potential alternatives. Current DLR implementations like TrueFidelity and Advanced Intelligent Clear-IQ Engine (AiCE) demonstrate promising outcomes. However, challenges in implementation and machine learning model reliability require further exploration. In conclusion, MAR algorithms enhance CT imaging quality by rectifying artifacts near metal implants, while DLR methods offer a promising path for radiation dose reduction and image refinement. Combining both approaches might pave the way for future CT imaging advancements.

Item Type:	Article (Review)
Subjects:	Q Science > QC Physics
Kulliyyahs/Centres/Divisions/Institutes (Can select more than one option. Press CONTROL button):	Kulliyyah of Medicine Kulliyyah of Medicine > Department of Radiology
Depositing User:	DR NOR AIN RABAIEE
Date Deposited:	02 Feb 2024 11:49
Last Modified:	20 Feb 2024 12:36
URI:	http://irep.iium.edu.my/id/eprint/110682

Actions (login required)

View Item