MRI adipose tissue segmentation and quantification in R (RAdipoSeg)

Area of Science:

• Biomedical imaging
• Medical physics
• Computational biology

Background:

• Excess adipose tissue, particularly visceral and subcutaneous depots, increases cardiovascular and metabolic risks.
• Magnetic Resonance Imaging (MRI) with fat suppression is crucial for accurate adipose tissue quantification.
• Existing methods for adipose tissue segmentation can be costly and complex.

Purpose of the Study:

• To develop and validate RAdipoSeg, a novel semi-automatic method for MRI adipose tissue segmentation.
• To assess the reliability and cost-effectiveness of RAdipoSeg compared to commercial software.
• To enable accurate fat quantification in both rodent models and human subjects.

Main Methods:

• Developed RAdipoSeg, a semi-automatic segmentation tool within the R statistical software.
• Acquired MRI images from wild-type mice on high/low-fat diets and 20 human subjects.
• Segmented 10 images per subject using RAdipoSeg and commercial software (SliceOmatic) for comparison.

Main Results:

• RAdipoSeg demonstrated comparable performance to commercial segmentation software.
• Mean Jaccard differences ranged from 10-29%, with relative volume differences below ±20%.
• High correlation (p<0.05) and acceptable bias noted in Bland-Altman analysis, manageable with method flexibility.

Conclusions:

• RAdipoSeg offers a reliable and low-cost solution for MRI-based fat segmentation.
• The method is suitable for adipose tissue quantification in studies involving both mice and humans.
• RAdipoSeg facilitates research into the metabolic risks associated with adipose tissue distribution.