To evaluate the application value of deep learning image reconstruction (DLIR) algorithm combined with pre-adaptive statistical iterative reconstruction-V (ASIR-V) in CT portal venography (CTPV) for overweight patients.

50 patients receiving abdominal enhanced CTPV in the Third Affiliated Hospital of Sun Yat-sen University from June to September 2021 were enrolled in this prospective study. Among them, 31 patients were male and 19 female, aged from 19 to 74 years, with a median age of 56 years and BMI>25 kg/m². The informed consents of all patients were obtained and the local ethical committee approval was received. The effective dose (ED) was compared between 50% pre-ASIR-V (turned on) and 0% pre-ASIR-V (turned off). The data of portal venous phase were thin-slice reconstructed with 4 algorithms 60%ASIR-V, 80%ASIR-V, DLIR-M (middle level) and DLIR-H (high level), respectively. The objective evaluation parameters of image quality included standard deviation (SD) of CT value of portal vein, signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR). The subjective assessment of reconstructed images was performed with double-blind method by two radiologists. The ED before and after turning on pre-ASIR-V was statistically compared by t test. The SD, SNR, CNR and other objective evaluation parameters of image quality among different algorithms were compared by one-way ANOVA. The subjective scores were compared by Kruskal-Wallis test. The consistency of subjective scores of image quality between two radiologists was analyzed by Kappa test.

The average ED was (11.1±1.4) and (7.6±1.1) mSv before and after turning on pre-ASIR-V, respectively, which was decreased by 32% (t=14.01, P<0.05). For the main portal vein, the SD was the lowest, the SNR and CNR were the highest in DLIR-H group (P<0.05). For the portal vein branches, the SD was the lowest and the SNR was the highest in 80%ASIR-V group, and the CNR was the highest in DLIR-H group (P<0.05). Among all the images reconstructed by different algorithms, the subjective score was the highest in DLIR-H group (P<0.05). The consistency of scores of two radiologists was comparatively high for the portal vein images reconstructed with 60%ASIR-V, 80%ASIR-V, DLIR-M and DLIR-H algorithms (κ=0.810, 0.556, 0.705, 0.676; P<0.05). Comparatively high consistency was also observed in the subjective scores for the images of portal vein system and portal vein branches (κ=0.661, 0.959; P<0.5).

In the CTPV for overweight patients, pre-ASIR-V can significantly reduce the ED. DLIR algorithm can significantly reduce the noise without changing the texture. Compared with ASIR-V algorithm, DLIR algorithm can obtain better portal vein phase images, especially the DLIR-H algorithm.