Effect of iterative reconstruction on variability and reproducibility of epicardial fat volume quantification by cardiac CT

Abstract Background The epicardial fat volume (EFV) measured by cardiac CT has emerged as an important parameter for understanding the pathophysiology of coronary atherosclerosis. Objective We investigated the variability and reproducibility of EFV measurements and evaluated the effect of model-based type iterative reconstruction (M-IR) on measurement results. Methods Non-contrast cardiac CT data (tube voltage 120-kVp, tube current time product 32 mAs) of 30 consecutive patients were reconstructed with filtered back projection (FBP), hybrid type iterative reconstruction (H-IR), and M-IR using a slice thickness of 3.0 mm. CT attenuation and image noise was measured for all reconstructions. Two observers independently quantified EFV using semi-automated software and interobserver agreement was evaluated. Results There was no significant difference in the CT attenuation of the ascending aorta among the three reconstructions. The mean image noise on FBP-, H-IR-, and M-IR images was 48.0 ± 7.9 HU, 29.6 ± 4.8 HU, and 9.3 ± 1.3 HU, respectively; there was a significant difference among all comparison combinations for the three reconstructions (p  3 [FBP], 153.8 ± 53.1 cm 3 [H-IR], and 134.0 ± 46.4 cm 3 [M-IR]). For all three reconstructions, interobserver correlations were excellent (r = 0.91 [FBP], 0.93 [H-IR], and 0.96 [M-IR]). Interobserver comparisons showed that the lowest Bland–Altman limit of agreement was with M-IR (mean difference 2.0 ± 4.9%, 95% limit of agreement, −24.0 to 28.0%) followed by H-IR (−2.6 ± 7.1%, −39.8 to 34.6%) and FBP (−0.2 ± 8.6%, −45.3- to 45.0%). Conclusion For the quantification of epicardial fat by cardiac CT, model-based iterative reconstruction can improve the image quality and lessen measurement variability.