Architectural overview of BioCompNet. (IMAGE)
Caption
BioCompNet: a dual-channel deep learning framework for automated body composition analysis from fat-water MRI sequences. (A) Schematic of the dual-channel 2-dimensional (2D) U-Net architecture used to automatically segment abdominal and thigh body composition components from MRI scans. Channel 1 inputs fat-sequence images, which provide enhanced contrast for visualizing adipose tissue, while Channel 2 inputs water-sequence images, which offer better visualization of muscles, bones, and vessels. The network structure is illustrated on the right side using color-coded blocks. The segmented body composition components are represented by different colors for visual differentiation. (B) Based on the segmentation outputs, a fully automated postprocessing algorithm was developed to quantify morphological features (e.g., volume, circumference, and cross-sectional area) for each component. In addition, a K-means clustering algorithm (k = 2) was applied to the core muscle (CM) and thigh muscle regions to identify intermuscular adipose tissue (IMAT). To assess the accuracy of the automated results produced by BioCompNet, a consistency analysis was performed against reference evaluations conducted by experienced physicians. AI, artificial intelligence; GT, ground truth; MRI, magnetic resonance imaging; VB, vertebral bone; PM, psoas muscle; SAT, subcutaneous adipose tissue; sSAT, superficial subcutaneous adipose tissue; dSAT, deep subcutaneous adipose tissue; IPAT, intraperitoneal adipose tissue; RPAT, retroperitoneal adipose tissue; ICC, intraclass correlation coefficient.
Credit
Jianyong Wei, Shanghai Jiao Tong University School of Medicine.
Usage Restrictions
News organizations may use or redistribute this image, with proper attribution, as part of news coverage of this paper only.
License
Licensed content