Repeatability of automated body composition measurement on low dose chest CT in male subjects

The objective was to determine the most repeatable of three automated body composition methods applied to baseline and short-term follow-up chest CT scans.

Areas of skeletal muscle and subcutaneous adipose tissue (SAT) were analyzed in a 1 mm slice close to the aortic arch in a subset of males from the NELSON lung cancer screening trial with baseline and 3-4 month repeat CT scan. We compared three pre-existing machine learning methods we call: truncated field of view (FOV), compensated FOV, and extended FOV, of which the last two can deal with non-overlapping FOV in scans. Repeatability was assessed using Bland-Altman plots and paired T-tests.

Of 562 males the median (interquartile range) age was 60.8 (56.3-64.8) years. Mean skeletal muscle areas were similar for truncated (212 cm²) and extended FOV (211 cm²), and slightly lower for compensated FOV (208 cm²) (p < 0.001). SAT areas were higher with extended FOV (156 cm²) compared to truncated (132 cm²) and compensated FOV (125 cm²) (p < 0.001). A small systematic longitudinal difference in skeletal muscle was observed for extended FOV (mean±SD 1.7 ± 17.3 cm2, p = 0.017). Limits of agreement for skeletal muscle area were -18.9% to 20.4% for truncated FOV, -11.1% to 11.6% for compensated FOV, and -16.7% to 18.2% for extended FOV. Corresponding values for SAT area were -37.3% to 38.3%,-30.2% to 29.3%, and -29.1% to 29.9%.

Extended FOV had the second-most repeatable measurements and was unaffected by FOV cutoff. Compensated FOV was most repeatable, but underestimated SAT.