CT Body Composition Accuracy Depends Heavily on Scanner Settings

CT-based body composition analysis (BCA) has become a cornerstone of clinical research into sarcopenia, obesity, and disease prognosis, largely because millions of patients already undergo abdominal CT, enabling opportunistic measurement of muscle and fat without additional scans. The method relies on Hounsfield unit (HU) thresholds — originally established at 120 kV — to segment skeletal muscle (−29 to +150 HU), visceral adipose tissue (VAT), and subcutaneous adipose tissue (SAT). Muscle quality is assessed via mean muscle attenuation, with lower values indicating fat infiltration (myosteatosis). Deep-learning segmentation has further automated these measurements, accelerating large-scale cohort studies. However, this review from Eunpyeong St. Mary's Hospital synthesizes evidence showing that virtually every technical parameter of CT acquisition and reconstruction introduces measurable variability into these metrics.

Contrast material is among the most clinically significant variables. Seven of ten studies in a scoping review reported significantly increased muscle attenuation after contrast injection, and post-contrast images generally show greater skeletal muscle area (SMA) and skeletal muscle density (SMD) regardless of injection protocol. One study documented a 7.6% average decrease in VAT area and a 5.4% increase in VAT attenuation on post-contrast images, while SAT and SMA changed only marginally (0.1–0.2%). Although pre- and post-contrast values correlate well, they are not interchangeable. A proposed correction of −7.5 HU applied to post-contrast muscle attenuation to approximate non-contrast values is not universally valid, and mixing contrast phases in research datasets introduces systematic bias.

The specific phase of contrast imaging matters independently. Across three- and four-phase CT protocols, SMA, normal-attenuation muscle area (NAMA), and SMD progressively increase from unenhanced through arterial and portal venous to delayed phases, while low-attenuation muscle area (LAMA) decreases — meaning fewer patients would be classified as having myosteatosis in later phases. VAT area decreases in delayed imaging, and when predefined obesity thresholds are applied, the prevalence of visceral obesity is significantly lower in later phases. One perfusion CT study confirmed a time-dependent increase in skeletal muscle index after contrast injection with simultaneous reductions in LAMA and adipose tissue index.

Tube voltage and current each introduce additional variability. At 80 kV versus 120 kV, SMD is significantly greater while LAMA is significantly lower in clinical portal-venous phase imaging, with a discrepancy in phantom studies (where LAMA increased at 80 kV) likely explained by the absence of contrast material in phantoms. Fat attenuation increases at higher voltages, but area-based adipose indices may not differ significantly, meaning the choice of metric matters for interpretation. Tube current reduction — used for dose savings — yielded similar SMA but lower muscle attenuation in phantom studies when cut to 10–50% of standard dose; a clinical comparison found significantly lower VAT and SAT areas and attenuation in low-dose images (mean 28.8 mAs vs. 161.9 mAs).

Reconstruction algorithm is another major source of variability covered in the review. Iterative reconstruction (IR) and deep-learning reconstruction (DLR) reduce noise compared with filtered back projection (FBP), but the degree of noise suppression affects HU histogram distributions and therefore attenuation-based metrics like SMD and LAMA. Slice thickness is particularly impactful: thicker slices increase partial-volume averaging, altering the apparent attenuation of small structures and muscle boundaries. Emerging technologies — dual-energy CT virtual monoenergetic images and photon-counting CT — offer potential advantages for standardization but introduce new sources of variability that remain understudied in BCA contexts. The review concludes that consistent reporting of all acquisition and reconstruction parameters is essential for reproducibility in both clinical practice and multicenter research.