Comparison of image quality and pancreatic ductal adenocarcinoma conspicuity between the low-kVp and dual-energy CT reconstructed with deep-learning image reconstruction algorithm

To compare the image quality and conspicuity of pancreatic ductal adenocarcinoma (PDAC) between the low-kVp and dual-energy pancreatic protocol CT reconstructed with deep-learning image reconstruction (DLIR).A cohort of 111 consecutive patients (median age, 72 years; 56 men) undergoing a pancreatic protocol CT were retrospectively analyzed. Among them, 58 patients underwent 80-kVp CT (80-kVp group), and 53 patients underwent dual-energy CT and reconstructed at 40-keV (40-keV group). The medium-strength level of DLIR were used in both groups. Quantitative measurements, qualitative image quality, PDAC conspicuity, and dose-length product (DLP) were compared between the two groups using Mann-Whitney U test.A total of 20 and 16 PDACs were found in the 80-kVp and 40-keV groups, respectively. CT numbers of the vasculatures and parenchymal organs (P .001 for all) and the background noise at both pancreatic and portal venous phases (P .001) were higher in the 40-keV group than in the 80-kVp group. The signal-to-noise ratio (SNR) of all anatomical structures (P .001-0.005), except for the liver in reviewer 2 (P =.47), and the tumor-to-pancreas contrast-to-noise ratio (CNR; P .001-0.01) were higher in the 40-keV group than in the 80-kVp group. No difference was found in the image quality at both phases (P =.30-0.90). PDAC conspicuity was better in the 40-keV group than in the 80-kVp group (P =.007-0.03). DLP at pancreatic (275 vs. 313 mGy*cm; P =.05) and portal venous phases (743 vs. 766 mGy*cm; P =.20) was comparable between the two groups.Under the same DLP, virtual monoenergetic images at 40-keV demonstrated higher SNR and tumor-to-pancreas CNR and better PDAC conspicuity compared to the 80-kVp setting.