Abstract
ObjectivesThe escalating global incidence of obesity, cardiometabolic disease and sarcopenia necessitates reliable body composition measurement tools. MRI-based assessment is the gold standard, with utility in both clinical and drug trial settings. This study aims to validate a new automated volumetric MRI method by comparing with manual ground truth, prior volumetric measurements, and against a new method for semi-automated single-slice area measurements.Methods4905 individuals from the UK Biobank with repeat whole-body Dixon MRI scans were selected. MRI data were processed automatically to derive new (1) volumetric and (2) single-slice area measurements at L3 vertebral level for visceral adipose tissue (VAT), subcutaneous adipose tissue (SAT), and abdominal skeletal muscle (SM). For comparison, prior volumetric measurements of VAT and SAT were included. A separate set of scans from 100 subjects was randomly selected and body composition volumes and areas were manually segmented as ground truth.ResultsThe new automated volumetric measurements were found to have excellent agreement with manual ground truth with no substantial bias (ICC ≥ 0.96, CoV ≤ 3.8%). In the cohort of 4905 individuals (49% male, mean age 62 years ± 8, BMI 26 kg/m2 ± 4), we confirmed that prior and new volumetric methods of VAT and SAT measurement were very strongly correlated (VAT: ρ = 0.99; SAT: ρ = 1; both p < 0.001). Single-slice L3 area measurements demonstrated very strong correlations with corresponding volumes (for VAT ρ = 0.97, for SAT ρ = 0.94, for SM ρ = 0.95, all p < 0.001), and remained excellent across sex, age, and cardiometabolic characteristics (median ρ = 0.95).ConclusionWe found robust correlations between manually segmented, automated volumetric, and semi-automated single-slice body composition methods. The interchangeability of these methods suggests that for each application the method should be selected according to practical considerations, operational differences and measurement granularity, rather than technical performance.