Single-slice MRI for body composition assessment: repeatability, reproducibility, and observer variability

Purpose The single-slice MRI at L3 vertebra offers an efficient way to assess body composition but the evidence on the reliability of this method is scarce. This study evaluates the accuracy and precision of this method for adipose and muscle tissue measurements. Methods The technical performance of single-slice (L3) MRI body composition measurements was assessed in a prospective study of 12 participants, focusing on scan-rescan repeatability, cross-scanner reproducibility, and analyst variability. Additionally, retrospective data from 36 participants were analyzed to evaluate inter-device and inter-observer (analyst vs. radiologist) variability across a wide range of scanners and body types. Blinded analyses were performed for visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT) cross-sectional areas (CSA), VAT and SAT indices, VAT/SAT ratio, skeletal muscle CSA, skeletal muscle index (SMI), psoas muscle CSA, and psoas muscle index. Results Single-slice MRI-based body composition measurements showed high scan-rescan repeatability (CoV: 1.5%–7.9%, ICC: 0.97–1.0), with low repeatability coefficients (RC) across key metrics, including 12 cm ² for SAT CSA, 15 cm ² for VAT CSA, 5 cm ² for skeletal muscle CSA, and 1.4 cm ² for psoas muscle CSA. Cross-scanner reproducibility was consistent (CoV: 2.3%–15%, ICC: 0.90–1.0). Inter- and intra-analyst variability was minimal (CoV: 0.5%–5.0%, ICC: 0.98–1.0). Analyst-radiologist comparisons showed near-perfect correlations (r = 0.97–1.00, p < 0.001) and excellent reliability (ICC: 0.96–1.0). Conclusion The results demonstrate that MRI-based single-slice method at the L3 vertebral level provides accurate, repeatable, and reproducible measurements of adipose and muscle tissue across a wide range of body types, consistent between trained analysts and experienced radiologists. These findings support the method’s accuracy and consistency for longitudinal assessments.