Deep learning image reconstruction enables virtual non-contrast to replace true non-contrast in chest spectral CT
XU Long
LI Xin
DANG Shan
YU Nan
JIA Yongjun
DUAN Haifeng
Abstract：Objective To explore the application value of deep learning image reconstruction(DLIR)in optimizing the image quality of virtual non-contrast(VNC)chest spectral CT.Methods Forty-five patients undergoing true non-contrast(TNC)and dual-phase contrast-enhanced spectral CT of the chest at the Affiliated Hospital of Shaanxi University of Chinese Medicine from June to October 2024 were prospectively enrolled.ASIR-V50%weighted reconstruction at 120 kVp-like settings served as the true non-contrast reference(TNC-AR50).Based on arterial and venous phase contrast data,four DLIR-reconstructed VNC groups(VP-VNC-DM,VP-VNC-DH,AP-VNC-DM,AP-VNC-DH).CT values,noise(SD),SNR,and CNR were measured for the aorta,subcutaneous fat,erector spinae muscles,and lesions across all five image sets(TNC-AR50+4 VNC sets).Objective metrics were compared using one-way ANOVA and Kruskal-Wallis tests.Two radiologists independently performed subjective blinded evaluations of overall image quality and lesion visibility using a 5-point Likert scale.Results In objective image quality assessment,the VP-VNC-DH group demonstrated superior quality compared to TNC-AR50,with no statistically significant differences in CT values among the five groups(P>0.05).The VP-VNC-DH group exhibited the lowest image noise and the highest SNR and CNR.In subjective evaluation,the VP-VNC-DH group received the highest image quality scores and performed best in lesion conspicuity.The total effective radiation dose for chest CT with and without the TNC scan was 9.40±0.41 mSv and 6.27±0.28 mSv,respectively.Omitting the TNC scan reduced the total radiation dose by approximately 33.3%.Conclusion In chest-enhanced CT examinations,VNC images reconstructed using DLIR(especially venous-phase DLIR-H)demonstrated significantly superior image quality compared to TNC images reconstructed using ASIR-V 50%,with good CT value consistency.It is recommended to use venous-phase high-level DLIR(DLIR-H)reconstruction for VNC images as an alternative to true non-contrast scans to effectively reduce radiation dose.
Keywords：deep learning reconstruction algorithmvirtual plain scanchest CTradiation dose
Publication Date：2026-01-20
Online Publishing Date：2026-02-04(First online date of this platform, not the publication date of the document)
Pages：6( 44-49 )