Cross-Modality Validation of Abdominal Fat and Muscle Segmentation: A Comparative Study of Dixon MR and CT Imaging

• Published in: IEEE International Conference on Big Data pp. 8762 - 8764
• Main Authors: Liu, Andrew, Heller, Andrew, Cohen, Gregg, Jones, Elizabeth C., Hsu, Li-Yueh
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15.12.2024
• Subjects: Abdomen; Accuracy; adipose tissue; Computed tomography; Correlation; Dixon; Fats; Linear regression; Magnetic resonance imaging; Muscles; segmentation; Software; Thresholding (Imaging)
• ISSN: 2573-2978
• DOI: 10.1109/BigData62323.2024.10826022

This study evaluates the agreement between Dixon-based MRI and CT in quantifying abdominal muscle and adipose tissue areas, aiming to establish MRI as an accurate, radiation-free alternative to the CT gold standard. Twenty subjects underwent abdominal CT and Dixon MRI on the same day, with matched axial images at L2-L3 and L4-L5 analyzed using semi-automatic software to contour boundaries, apply intensity thresholding, followed by manual refinement of the fat and muscle masks. Bland-Altman plots and linear regression analyses revealed strong agreement between MRI and CT for muscle and subcutaneous adipose tissue (SAT) areas, with mean differences of -0.02 cm 2 and - 1.13 cm 2 and limits of agreement within ±20.46 cm 2 and ±34.71 cm 2 , respectively, while visceral adipose tissue (VAT) showed larger discrepancies, likely due to compression of the abdomen during MRI, with a mean difference of -18.58 cm 2 and a limit of agreement of 20.34 cm 2 . Linear regression confirmed strong correlations with R 2 values of 0.89 for muscle, 0.98 for SAT, and 0.93 for VAT. These findings support MRI as a precise and radiation-free alternative for body composition analysis, particularly for muscle and SAT.