Related Experiment Video
Updated: Aug 6, 2026
08:52
3D Ultrasound Imaging: Fast and Cost-effective Morphometry of Musculoskeletal Tissue
Published on: November 27, 2017
Development of a Deep Learning Model for Automated Measurement of Skeletal Muscle Volume in 18F-FDG PET/CT
Ryusuke Nakamoto1, Koji Fujimoto2, Ryo Sakamoto3
1Preemptive Medicine and Lifestyle Related Disease Research Center, Kyoto University Hospital, Kyoto, Kyoto, Japan. inabook@kuhp.kyoto-u.ac.jp.
Journal of Imaging Informatics in Medicine
|July 20, 2026
Summary
A new deep learning method accurately measures trunk muscle volume using PET/CT scans. This automated 3D volumetry is more robust than 2D methods for body composition analysis.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate body composition analysis, particularly muscle mass, is crucial for health assessment.
- Traditional methods for trunk muscle volumetry can be time-consuming or lack precision.
- Whole-body CT scans from PET/CT offer a potential data source for advanced analysis.
Purpose of the Study:
- To develop and validate a deep learning-based automated method for trunk muscle volumetry using whole-body CT data from PET/CT scans.
- To compare the performance of this automated 3D volumetry against bioelectrical impedance analysis (BIA) and conventional 2D cross-sectional area (CSA) measurements.
- To assess the robustness of the automated method across different sexes.
Main Methods:
- A retrospective study utilizing an nnU-Net-based segmentation model trained on 20 manually annotated PET/CT datasets.
- Segmentation of trunk muscles was performed on 209 individuals who also underwent BIA.
- Model performance was evaluated using Dice similarity coefficients (DSCs), and correlations between muscle mass estimates (3D volume vs. 2D L3-CSA) and BIA-derived mass were analyzed using Pearson's correlation coefficients.
Main Results:
- The automated segmentation model achieved a high Dice similarity coefficient (DSC) of 0.991.
- Automated 3D muscle volume showed a significantly stronger correlation with BIA-derived muscle mass (r=0.961) compared to conventional 2D L3-CSA (r=0.912).
- The 3D volumetry method maintained high correlations in both sexes, whereas 2D L3-CSA performance was significantly reduced in men.
Conclusions:
- Automated 3D trunk muscle volumetry using whole-body CT from PET/CT provides a significantly more robust and accurate assessment of muscle mass than conventional 2D metrics.
- This deep learning framework enables scalable and precise body composition analysis within existing PET/CT workflows without additional radiation exposure.
- The method offers a valuable tool for clinical research and practice, particularly for evaluating muscle health and variations in body composition.
