Background: The effect of differing post-contrast phases on CT body composition measurements is not yet known. Methods: A fully automated AI-based body composition analysis using DAFS was performed on a retrospective cohort of 278 subjects undergoing pre-treatment triple-phase CT for pancreatic intraductal papillary mucinous neoplasm. The CT contrast phases included noncontrast (NON), arterial (ART), and venous (VEN) phases. The software selected a single axial CT image at mid-L3 on each phase for body compartment segmentation. The areas (cm2) were calculated for skeletal muscle (SM), intermuscular adipose tissue (IMAT), visceral adipose tissue (VAT), and subcutaneous adipose tissue (SAT). The mean Hounsfield units of skeletal muscle (SMHU) within the segmented regions were calculated. Bland–Altman and Chi Square analyses were performed. Results: SM-NON had a lower percentage of bias [LOA] than SM-ART, −0.7 [−7.6, 6.2], and SM-VEN, −0.3 [−7.6, 7.0]; VAT-NON had a higher percentage of bias than ART, 3.4 [−18.2, 25.0], and VEN, 5.8 [−15.0, 26.6]; and this value was lower for SAT-NON than ART, −0.4 [−14.9, 14.2], and VEN, −0.5 [−14.3, 13.4];] and higher for IMAT-NON than ART, 5.9 [−17.9, 29.7], and VEN, 9.5 [−17.0, 36.1]. The bias in SMHU NON [LOA] was lower than that in ART, −3.8 HU [−9.8, 2.1], and VEN, −7.8 HU [−14.8, −0.8]. Conclusions: IV contrast affects the voxel HU of fat and muscle, impacting CT analysis of body composition. We noted a relatively smaller bias in the SM, VAT, and SAT areas across the contrast phases. However, SMHU and IMAT experienced larger bias. During threshold risk stratification for CT-based measurements of SMHU and IMAT, the IV contrast phase should be taken into consideration.
Read full abstract