Objective To investigate the feasibility of deep learning image reconstruction(DLIR)algorithm in impro-ving the image quality of thin-slice contrast-enhanced CT(CECT)scan and the lesion conspicuity of pancreatic ductal ade-nocarcinoma(PDAC)as compared to filtered back projection(FBP)and hybrid model-based adaptive statistical iterative reconstruction(ASIR-V).Methods A retrospective analysis was performed on 44 patients with pathologically confirmed PDAC who underwent preoperative dual-phase CECT scanning.The raw data from portal-venous images were reconstructed using FBP,60％ASIR-V,and DLIR(low,medium,and high strength levels)algorithms at a slice thickness of 0.625 mm.Analysis of variance and Friedman test were used to compare objective(noise in HU,image texture,low-contrast res-olution,high-contrast resolution)and subjective(image noise,sharpness,overall image quality and tumor conspicuity)in-dicators between groups.Results The image texture and low-contrast resolution of DLIR were comparable to FBP but better than 60％ASIR-V,and the noise values of DLIR were lower compared to FBP and 60％ASIR-V(all P＜0.05).All subjective indicators of DLIR were better than or similar to FBP and 60％ASIR-V.With the increase of DLIR strength lev-el,the high-contrast resolution and sharpness scores did not change significantly,the low-contrast resolution and noise de-creased,the image texture is blurred,and the overall image quality and tumor conspicuity increased.Conclusion DLIR can significantly improve the image quality of thin-slice enhanced CT images and enhance the lesion conspicuity of PDAC.
NETL
NSTL
万方数据
