Self-supervised multi-organ segmentation for aging pattern exploration

Understanding aging patterns in abdominal organs requires a comprehensive framework that integrates morphological and compositional analyses. This study pioneers a complete technical pipeline from single-energy CT (SECT) scans to aging pattern exploration. We aim to develop this framework by combining high-accuracy multi-organ segmentation with basis material decomposition, enabling simultaneous analysis of structural changes and compositional variations in abdominal organs. We constructed a multi-center dataset (BVTAMOS+) comprising 6487 abdominal CT volumes, including 170 volumes with voxel-level annotations, covering 2365 annotated organs. The core of this framework is the Anatomical Information Guided Multi-Scale Parallel U-Net (AIG-MSPU), which incorporates Anatomical Probability Density Maps (APDM) and Adaptive Multi-Scale Fusion (AMSF) to enhance segmentation accuracy and robustness across diverse organ structures. Additionally, a two-stage material decomposition approach was used to extract compositional information from SECT scans. The proposed segmentation method achieved an average Dice Similarity Coefficient (DSC) of 86.15% and a 95% Hausdorff Distance (HD95) of 6.18 mm. Based on the developed framework, we analyzed 1074 SECT scans and observed age-associated changes in liver and kidney morphology and proxy composition indices. Segmented regression with prespecified knots at 60 and 70 years quantified age-segment specific slopes, with volume decline becoming more evident beyond age 60. Furthermore, kidney atrophy showed dimension-dependent patterns, whereas liver shrinkage appeared more uniform across dimensions. This framework enables large-scale joint assessment of abdominal organ morphology and proxy composition from routine SECT, and provides quantitative evidence for aging-related trends at the population level, serving as a basis for external validation and clinical translation.