Artificial Intelligence Iterative Reconstruction in Multiphase Abdominopelvic CT: Added Value for Anatomic Evaluation of Uterine Artery.
Authors
Affiliations (2)
Affiliations (2)
- Department of Radiology, The First Affiliated Hospital of Shandong First Medical University & Shandong Provincial Qianfoshan Hospital, Jinan, Shandong.
- United Imaging Healthcare, Shanghai, China.
Abstract
To assess the anatomic evaluation performance of the uterine artery (UA) on routine multiphase abdominopelvic CT images, and to investigate whether artificial intelligence iterative reconstruction (AIIR) can improve this performance to a level comparable to that on CTA images. This retrospective study included 50 patients who underwent multiphase abdominopelvic CT. Aarterial phase (AP) images were reconstructed with AIIR (group B) in addition to the routine hybrid iterative reconstruction (HIR) (group A). Another 50 patients who underwent routine abdominal aorta CTA were retrospectively identified, and their images were reconstructed with HIR (group C). The anatomic evaluation performance of UA in terms of the interpretability of origins, branching pattern, and lumen diameter was compared across the 3 image sets. The objective image quality was also compared. A total of 100 patients (age 18 to 65 y) were investigated, and 300 UAs were evaluated (150 images × 2 UAs). Both group B and group C outperformed group A in terms of the proportion of UAs with identifiable origins (90% and 88% vs. 68%), classifiable branching pattern (89% and 87% vs. 71%), and totally measurable diameter (63% and 67% vs. 33%) (all P<0.017). No significant differences were found between groups B and C for any of these proportions (all P>0.017). The measured diameters exhibited excellent intraobserver agreement with ICC values of 0.90, 0.93, and 0.98 for groups A, B, and C, respectively. The UA attenuation was highest in group C, followed by group B and group A (all P<0.017). Group B exhibited the lowest noise in all scenarios (all P<0.017). No significant difference was found in the CNR of UA between group B and group C (all P>0.017). Sensitivity analysis excluding hypertensive patients confirmed the stability of the results. Anatomic evaluation performance of UA on routine multiphase abdominopelvic CT images is suboptimal. However, when AIIR is applied, both the interpretability of anatomic features and the image quality of UA are comparable to those obtained with dedicated CTA.