Energy Spectral Single-energy Technique Based on Deep Learning Image Reconstruction:Study on Image Quality of Thoracic Aorta under Low Contrast Agent Flow Rate
YE Xiongxin
LIU Yuanfen
TANG Borong
CHEN Yilin
ZHENG Wanyi
XUE Liwei
ZHANG Xiaoyong
Abstract：Objective:To investigate the value of combining a deep learning image reconstruction algorithm and an energy spectral single-energy technique to improve the image quality of the thoracic aorta with a low contrast agent flow rate.Materials and Methods:The imaging data of 50 patients with thoracic aorta energy spectral CTA scans with contrast agent flow rate≤1.5 mL/s from January 2016 to December 2023 at Fujian Medical University Union Hospital were retrospectively analyzed and whose thoracic aorta enhancement was insufficient(thoracic aorta CT value<250 HU)on 120 kVp-like images.ASIR-V and two deep-learning image reconstructions(DLIR-M and DLIR-H)were performed on kVp-like images,40 keV,50 keV,and 60 keV single-energy images.The objective image quality parameters(thoracic aorta CT value,noise,SNR,CNR,and BHA)were compared with the subjective image quality scores.Images with thoracic aorta CT value≥250HU and subjective score≥3 were defined as meeting the diagnostic requirements.Results:CT values were 40 keV>50 keV>60 keV>120 kVp-like images.There was no statistically significant difference in the thoracic aortic CT values between the different reconstruction algorithms for the same type/energy level.SD,SNR,CNR,and BHA values were 40 keV>50 keV>60 keV>120 kVp-like images,respectively,and SD and BHA values were ASIR-V40%>DLIR-M>DLIR-H.The SNR and CNR of all the DLIR images(DLIR-M/H)at different energy levels were higher than those of the ASIR-V images.For subjective scoring,at the same energy level,DLIR-H>DLIR-M>ASIR-V,and under the same reconstruction algorithm:40 keV>50 keV>60 keV>120 kVp-like.All differences were statistically significant.All cases could obtain successful diagnostic images through 40 keV-DLIR-H.Conclusion:Spectral single-energy images combined with deep learning reconstruction algorithms can provide objective parameters that meet the diagnostic needs of thoracic aorta CT images with a poor enhancement effect under a low contrast agent flow rate while significantly improving the overall image quality.
Keywords：deep learning image reconstructionsingle-energythoracic aortabeam hardening artefactimage quality
Publication Date：2024-11-28
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：9( 683-691 )