Automated Abdominal Adipose Tissue Segmentation and Quantification in Children using Free-Breathing Dixon MRI and 3D Neural Networks

Motivation: The volume and fat content of subcutaneous and visceral adipose tissue (SAT, VAT) are associated with risk for cardiometabolic diseases. Although limited studies automatically segmented SAT, VAT on MRI in children, they targeted narrow age ranges and used conventional sequences that are sensitive to motion artifacts. Goal(s): To automatically segment and quantify the volume and fat fraction of abdominal SAT, VAT in children using motion-robust free-breathing radial Dixon MRI and state-of-the-art 3D neural networks. Approach: We enrolled 134 children (6-18 years), then trained 3D nnU-Net and Swin U-Net Transformer. Results: Both networks yielded accurate volume and fat fraction quantification with promising segmentation performance. Impact: We developed 3D neural networks to automatically segment and quantify abdominal subcutaneous and visceral adipose tissue (SAT, VAT) characteristics in children (6-18 years) using free-breathing MRI. These methods can be used to study risk factors for cardiometabolic diseases in children.