Automated Deep Learning-Based Segmentation of Abdominal Adipose Tissue on Dixon MRI in Adolescents: A Prospective Population-Based Study.

Background: Prevalence of childhood obesity has increased significantly worldwide, highlighting a need for accurate noninvasive quantification of body fat distribution in children. Objective: To develop and test an automated deep learning method for subcutaneous adipose tissue (SAT) and visceral adipose tissue (VAT) segmentation using Dixon MRI acquisitions in adolescents. Methods: This study was embedded within the Generation R Study, a prospective population-based cohort study in Rotterdam, the Netherlands. The present study included 2989 children (mean age, 13.5 years; 1432 boys, 1557 girls) who underwent investigational whole-body Dixon MRI after reaching age 13 years, during the Generation R Study's follow-up phase. A competitive dense fully convolutional network (2D-CDFNet) was trained from scratch to segment abdominal SAT and VAT using Dixon-based images. The model underwent training, validation, and testing in 62, 8, and 15 children, respectively, selected by stratified random sampling, using manual segmentations as reference. Segmentation performance was assessed using Dice similarity coefficient and volumetric similarity. Two observers independently performed subjective visual assessments of automated segmentations in 504 children, selected by stratified random sampling, scoring undersegmentation and oversegmentation on 0-3 scales (3=nearly perfect segmentation). In 2820 children with complete data, Spearman correlation coefficients were computed between MRI measurements with BMI and dual-energy X-ray absorption (DEXA)-based measurements. The model is publicly available: https://gitlab.com/radiology/msk/genr/abdomen/cdfnet Results: In the test dataset, Dice similarity coefficient and volumetric similarity were, for SAT, 0.94±0.03 and 0.98±0.01, and for VAT, 0.85±0.05 and 0.92±0.04. The two observers assigned score of 3 for SAT in 94% and 93% for undersegmentation proportion and 99% and 99% for oversegmentation proportion, and score of 3 for VAT in 99% and 99% for undersegmentation proportion and 95% and 97% for oversegmentation proportion. Correlations with SAT and VAT were 0.808 and 0.698 for BMI, and 0.941 and 0.801 for DEXA-derived fat mass. Conclusion: We trained and evaluated the 2D-CDFNet model on Dixon MRI in adolescents. Quantitative and qualitative measures of automated SAT and VAT segmentations indicated strong model performance. Clinical Impact: The automated model may facilitate largescale studies in adolescents investigating abdominal fat distribution on MRI, as well as associations of fat distribution with clinical outcomes.