Adaptive deep learning for quantification and spatial distribution of abdominal adipose tissue from magnetic resonance imaging proton density fat fraction in adults with a body mass index ≥24 kg/m<sup>2</sup>.
Authors
Affiliations (3)
Affiliations (3)
- College of Medical Imaging, Shanghai University of Medicine & Health Sciences, Shanghai, China.
- Department of Radiology, Shanghai East Hospital, Tongji University, Shanghai, China.
- Shanghai Key Laboratory of Magnetic Resonance, East China Normal University, Shanghai, China.
Abstract
Abdominal adipose tissue volume and spatial distribution are closely associated with a variety of metabolic diseases and reflect individual metabolic risk and health status. This study aimed to evaluate the feasibility of an adaptive deep learning framework for the automated volumetric quantification and spatial mapping of visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT) on magnetic resonance imaging (MRI) proton density fat fraction (PDFF) in adults with a body mass index (BMI) ≥24 kg/m<sup>2</sup>. In this single-center retrospective study, a total of 238 abdominal MRI PDFF datasets from adults with a BMI ≥24 kg/m<sup>2</sup>, acquired with a 3.0-T mDixon sequence, were analyzed. A no new U-net (nnU-Net)-based adaptive deep learning model was trained to automatically segment abdominal VAT and SAT and to generate volumetric measurements across all slices. Spatial adipose distribution maps were constructed and stratified by sex and age. The adaptive deep learning-based segmentation model achieved high segmentation performance on the internal independent test set, with a mean Dice coefficient >0.966 and a mean relative volume error of -1.0335%. The framework enabled rapid and automated abdominal fat quantification, requiring approximately 40 seconds per case for automatic segmentation, much shorter than the approximately 1-2 hours per case for manual segmentation. Regional spatial distribution analysis showed pronounced heterogeneous distribution patterns of VAT and SAT across different abdominal anatomical regions, with distinct differences in spatial distribution patterns across sex and age groups. Specifically, VAT volume was significantly higher in men than in women (P<0.001). In the age-stratified analysis, SAT volume differed significantly across age groups (P<0.001), whereas the difference in VAT volume did not reach statistical significance (P=0.096). In the BMI category-stratified analysis, both SAT and VAT volumes differed significantly across BMI categories (both P<0.001). Within the same BMI categories, VAT showed greater relative variability than did SAT, with coefficients of variation ranging from 36.93% to 56.67% and from 23.52% to 33.30%, respectively. An adaptive deep learning framework based on nnU-Net enables automated quantification and spatial distribution characterization of abdominal adipose tissue in overweight and obese Chinese individuals and may thus serve as methodological platform for subsequent multicenter external validation and clinical translation studies.