Utility of Dark-Blood Dual-Energy CT Images for Predicting Vascular Involvement and R0 Resection in Patients With Pancreatic Cancer

BACKGROUND. Current CT criteria for assessing vascular involvement by pancreatic ductal adenocarcinoma (PDAC) use circumferential contact as an indirect indicator. Dark-blood images derived from dual-energy CT (DECT) provide high lumen-to-wall contrast and may aid assessment. OBJECTIVE. The purpose of this study was to compare the diagnostic performance of 55-keV virtual monoenergetic images (VMIs) assessed using NCCN criteria with that of dark-blood images assessed using wall-based criteria for predicting vascular involvement and surgical resection that achieves microscopically negative margins (i.e., R0 resection) in patients with PDAC who undergo contrast-enhanced DECT. METHODS. This retrospective study included 109 patients (mean age, 62.6 +/- 8.8 [SD] years; 66 men, 43 women) with histologically confirmed PDAC who underwent pancreatic parenchymal and portal venous phase DECT within 4 weeks before surgery (including PDAC resection in 73 patients) between July 2020 and June 2022. Dark-blood images were derived using a two-material decomposition algorithm. Two radiologists independently reviewed 55-keV VMIs and dark-blood images in separate sessions to evaluate celiac artery, common hepatic artery, superior mesenteric artery, portal vein, and superior mesenteric vein involvement; a third radiologist resolved discrepancies. On 55keV VMIs, vessel relationships were classified as no contact, abutment (= 180 degrees contact), or encasement (> 180 degrees contact). On dark-blood images, vessel walls were categorized as intact circumferentially, irregular, or discontinuous. Tumor resectability status was classified on the basis of vessel relationships. Surgical observation served as the reference for vascular involvement. Margin status was determined for resected tumors. RESULTS. Across the five vessels, for predicting vascular involvement, abutment or encasement on 55-keV VMIs had sensitivity of 100.0% (all vessels) and specificity of 66.2-92.9%, and an irregular or discontinuous wall on dark-blood images had sensitivity of 80.0-100.0% and specificity of 88.2-98.0%. Specificity was higher for an irregular or discontinuous wall than for abutment or encasement for all vessels (all p <.05); sensitivity was not different for any vessel (all p >.05). Resectable disease classified by darkblood images, compared with resectable disease classified by 55-keV VMIs, showed no difference in sensitivity (89.5% vs 78.9%, p =.33) but showed higher specificity (75.9% vs 59.3%, p =.01) for predicting R0 resection. CONCLUSION. Dark-blood images showed higher diagnostic performance than 55keV VMIs for predicting vascular involvement and R0 resection in patients with PDAC. CLINICAL IMPACT. Dark-blood images may aid decisions regarding neoadjuvant therapy and surgical planning for PDAC.