Automated Muscle and Fat Segmentation in Computed Tomography for Comprehensive Body Composition Analysis

TL;DR

This work presents a publicly accessible end-to-end CT body composition segmentation framework that jointly segments skeletal muscle, SAT, VAT, and muscular fat across the chest–pelvis region and automatically computes clinically relevant metrics in both 2D at $L3$ and 3D across $T12$–$L4$. It evaluates nine segmentation architectures (including nnU-Net variants and foundation-model fine-tuning) on internal and SAROS datasets, selecting nnU-Net ResEnc XL for final benchmarking and demonstrating Dice scores exceeding 89% and mean relative absolute errors under 10% across metrics. The model achieves strong agreement with manual annotations and outperforms benchmark tools like TotalSegmentator and Enhanced Segmentation by notable margins on SAT and skeletal muscle segmentation. Demographic analyses reveal age- and sex-related differences in muscle and fat distribution, supporting the framework’s utility for prognostic and nutritional assessments. The work provides publicly available code and weights, enabling broad adoption and standardized body composition analysis in research and clinical contexts, while outlining avenues for extending to additional body regions and improving muscular fat labeling. $L3$ and $T12$-$L4$ based measurements, together with robust 2D/3D metrics, offer a scalable, reproducible platform for investigating body composition in diverse populations.

Abstract

Body composition assessment using CT images can potentially be used for a number of clinical applications, including the prognostication of cardiovascular outcomes, evaluation of metabolic health, monitoring of disease progression, assessment of nutritional status, prediction of treatment response in oncology, and risk stratification for surgical and critical care outcomes. While multiple groups have developed in-house segmentation tools for this analysis, there are very limited publicly available tools that could be consistently used across different applications. To mitigate this gap, we present a publicly accessible, end-to-end segmentation and feature calculation model specifically for CT body composition analysis. Our model performs segmentation of skeletal muscle, subcutaneous adipose tissue (SAT), and visceral adipose tissue (VAT) across the chest, abdomen, and pelvis area in axial CT images. It also provides various body composition metrics, including muscle density, visceral-to-subcutaneous fat (VAT/SAT) ratio, muscle area/volume, and skeletal muscle index (SMI), supporting both 2D and 3D assessments. To evaluate the model, the segmentation was applied to both internal and external datasets, with body composition metrics analyzed across different age, sex, and race groups. The model achieved high dice coefficients on both internal and external datasets, exceeding 89% for skeletal muscle, SAT, and VAT segmentation. The model outperforms the benchmark by 2.10% on skeletal muscle and 8.6% on SAT compared to the manual annotations given by the publicly available dataset. Body composition metrics show mean relative absolute errors (MRAEs) under 10% for all measures. Our model with weights is publicly available at https://github.com/mazurowski-lab/CT-Muscle-and-Fat-Segmentation.git.

Figures (13)

• Figure 1: Quantitative Evaluation of Vertebrae Detection by TotalSegmentator: The box plot in the left panel shows the slice distance between human-selected and automatically selected slices across different slice spacings. The right panel displays the percentage difference in body composition metrics caused by slice selection mismatch.
• Figure 2: Qualitative evaluation of our segmentation model: Figure shows segmentation results of the abdominal L3 slice. In the segmentation, dark blue shows skeletal muscle, light blue SAT, yellow VAT, and maroon muscular fat.
• Figure 3: $\mathbf{R^2}$ correlation plots: for muscle (top row) and SAT (bottom row) across three evaluation ranges: L3 slice, T12--L4 range, and the All Slices setting.
• Figure 4: Representative failure cases of our segmentation model: Each row shows one case, with the top two from the internal dataset and the bottom two from the external dataset. From left to right: original CT slice, zoomed-in view, ground truth segmentation, model prediction, and corresponding Dice scores. In the Ground Truth, green shows skeletal muscle, purple SAT, orange VAT, and yellow muscular fat. In the Pred, dark blue shows skeletal muscle, light blue SAT, yellow VAT, and maroon muscular fat.
• Figure 5: Body composition metrics vs. age categories.