Super-resolution deep learning reconstruction improves the depiction of peripancreatic arteries and image quality in pancreatic cancer CT.

Authors

Abstract

Accurate identification of peripancreatic arteries in pancreatic ductal adenocarcinoma (PDAC) is essential for determining surgical strategy and prognosis. This study compared super-resolution deep learning reconstruction (SR-DLR) with normal-resolution deep learning reconstruction (NR-DLR) and hybrid iterative reconstruction (HIR) in the depiction of peripancreatic arteries in patients with PDAC undergoing pancreatic dynamic CT. This retrospective study included 40 patients (mean age, 70.8 ± 9.3 years) with pathologically confirmed PDAC who underwent pancreatic dynamic CT. Images were reconstructed using SR-DLR, NR-DLR, and HIR. In quantitative image analyses, CT attenuation, image noise, contrast-to-noise ratio (CNR), edge rise slope (ERS), and edge rise distance (ERD) were assessed. In qualitative image analyses, five radiologists independently evaluated arterial depiction of major arteries and small peripancreatic arteries, tumor-vessel interface, image noise, image sharpness, diagnostic acceptability, and tumor conspicuity. Statistical analyses were performed using paired t-tests and the Wilcoxon signed-rank test with Bonferroni correction. SR-DLR showed higher CT attenuation, lower image noise, and higher CNR than NR-DLR and HIR (all p < 0.001). For both the superior mesenteric artery and pancreatic parenchyma, ERS was higher with SR-DLR than with NR-DLR and HIR (all p ≤ 0.002), whereas ERD was lower than with NR-DLR and HIR (all p ≤ 0.001). In qualitative image analyses, SR-DLR achieved the highest scores for small peripancreatic arteries and all qualitative items except for the depiction of major peripancreatic arteries (p < 0.001). SR-DLR improved quantitative image quality and enhanced the qualitative depiction of small peripancreatic arteries and the tumor-vessel interface in pancreatic dynamic CT for patients with PDAC.

Tags

Topics