Evaluation of CT-based SUV normalization

The purpose of this study was to determine patients' lean body mass (LBM) and lean tissue (LT) mass using a computed tomography (CT)-based method, and to compare standardized uptake value (SUV) normalized by these parameters to conventionally normalized SUVs. Head-to-toe positron emission tomography (PET)/CT examinations were retrospectively retrieved and semi-automatically segmented into tissue types based on thresholding of CT Hounsfield units (HU). The following HU ranges were used for determination of CT-estimated LBM and LT (LBMCT and LTCT): -180 to -7 for adipose tissue (AT), -6 to 142 for LT, and 143 to 3010 for bone tissue (BT). Formula-estimated LBMs were calculated using formulas of James (1976 Research on Obesity: a Report of the DHSS/MRC Group (London: HMSO)) and Janmahasatian et al (2005 Clin. Pharmacokinet. 44 1051-65), and body surface area (BSA) was calculated using the DuBois formula (Dubois and Dubois 1989 Nutrition 5 303-11). The CT segmentation method was validated by comparing total patient body weight (BW) to CT-estimated BW (BWCT). LBMCT was compared to formula-based estimates (LBMJames and LBMJanma). SUVs in two healthy reference tissues, liver and mediastinum, were normalized for the aforementioned parameters and compared to each other in terms of variability and dependence on normalization factors and BW. Comparison of actual BW to BWCT shows a non-significant difference of 0.8 kg. LBMJames estimates are significantly higher than LBMJanma with differences of 4.7 kg for female and 1.0 kg for male patients. Formula-based LBM estimates do not significantly differ from LBMCT, neither for men nor for women. The coefficient of variation (CV) of SUV normalized for LBMJames (SUVLBM-James) (12.3%) was significantly reduced in liver compared to SUVBW (15.4%). All SUV variances in mediastinum were significantly reduced (CVs were 11.1-12.2%) compared to SUVBW (15.5%), except SUVBSA (15.2%). Only SUVBW and SUVLBM-James show independence from normalization factors. LBMJames seems to be the only advantageous SUV normalization. No advantage of other SUV normalizations over BW could be demonstrated.status: publishe