Effect of Image Reconstruction Algorithms in Pericoronary Adipose Tissue Attenuation obtained with Coronary CT Angiography

Background/Introduction Pericoronary adipose tissue (PCAT) attenuation measured at coronary CT angiography (CCTA) can detect vascular inflammation and is a useful diagnostic and prognostic marker. Modern iterative and artificial-intelligence-based CCTA reconstruction methods can affect image noise and overall image quality. We aimed to investigate the potential influence of different CCTA image reconstruction algorithms on PCAT attenuation measurements. Methods Patients referred for clinically indicated CCTA between March 2015 and June 2016 were retrospectively included. Datasets were reconstructed with filtered back projection (FBP), with adaptive statistical iterative reconstruction-Veo (ASiR-V) and with deep-learning image reconstruction (DLIR). In all three reconstructions, mean PCAT was evaluated in the three major epicardial coronary arteries using dedicated software (QAngio CT Research Edition version 3.2.0.13, Medis Medical Imaging Systems) and was defined as the area with an attenuation between − 30 and − 190 Hounsfield Units (HU) within a radial distance from the outer vessel wall equivalent to the diameter of the vessel. The proximal 40 mm segments of the left anterior descending artery (LAD) and left circumflex artery (LCx) and the proximal 10–50 mm segment of the right coronary artery (RCA) were analyzed. Using FBP (PCATFBP) as reference standard, correlation and agreement of PCAT estimates using ASIR-V (PCATASIR-V) and DLIR (PCATDLIHR) reconstructions were assessed. Results In 39 patients (mean age 57.9±11.2 years, 90% men), 117 arteries were analyzed. Mean PCATFBP attenuation of the overall population was -79.9 ±9.4 HU and did not differ significantly compared to PCATASIR-V and to PCATDLIR (-80.9±8.9 HU and -78.5±9.5 HU, respectively; pANOVA=0.123). Correlation was excellent for all comparisons [PCATFBP vs. PCATASIR-V; r=0.99, (y=4.68+1.046x, p<0.001), PCATFBP vs. PCATDLIR; r=0.99 (y=-3.57+0.973x, p<0.001)]. Intraclass correlation coefficient was excellent for all reconstructions [PCATFBP vs. PCATASIR-V; 0.99 (95% confidence interval: 0.98-0.99), PCATFBP vs. PCATDLIR; 0.98 (95% confidence interval: 0.98-0.99), all p<0.0001]. Agreement with Bland-Altman analysis was high with small mean absolute differences and relatively narrow limits of agreement [PCATFBP vs. PCATASIR-V; Bias: 1.0 HU (95% limits of agreement: -1.4 – 3.3), PCATFBP vs. PCATDLIR; Bias: -1.5HU (95% limits of agreement: -4.4 – 1.4)]. Conclusions CCTA-derived PCAT measurements using advanced iterative image reconstruction– and deep-learning image reconstruction–algorithms have excellent correlation and high agreement compared to conventional image reconstruction algorithms and could be potentially interchangeably used.