How Deep Learning Image Reconstruction (DLIR) affects the optimization of image quality and dose reduction on Computed Tomography

Abstract

The aim of this study is to investigate the influence of deep learning-based reconstruction (DLIR) on image quality across varying dose levels within a Chest- Abdomen-Pelvis (CAP) protocol using a 512-slice CT scanner and an advanced anthropomorphic phantom. Comparative analysis between DLIR, Adaptive Statistical Iterative Reconstruction (ASIR-V), and conventional Filtered BackProjection (FBP) reconstructions was conducted at normal, low, and ultra-low dose levels. The CT scanner employed in this experiment is the Revolution APEX by GE HealthCare (Waukesha, WI, USA). The experiment involves the use of a dedicated CT whole-body phantom, the PBU-60 by Kyoto Kagaku. A quantitative analysis was conducted, comparing the FBP Normal Dose (ND) and various reconstruction algorithms across three distinct dose levels (normal, low and ultra-low dose) and chest/abdomen/pelvis regions. Furthermore, an additional quantitative assessment was included, using ASIR-V60% as a reference due to its widespread utilization, between ASIR-V90% and DLIR-H. Also, a qualitative analysis performed to evaluate the general image quality and overall contrast of ASIR-V60%, ASIR-V90% and DLIR-H. The evaluation was carried out in terms of Signal-to-Noise Ratio (SNR) and Contrastto- Noise Ratio (CNR). The results highlight the feasibility of a low-dose protocol and suggest the potential introduction of an experimental ultra-low-dose protocol for CAP. The proposed implementation relies on the use of a deep-learning-based image reconstruction algorithm, which aims to maintain image quality and contrast levels comparable to those typically observed with conventional reconstruction algorithms used in regular and low-dose protocols.