# Age- and BMI-Dependent Psoas and Gluteus Muscle Mass in 27,805 Participants of the Population-Based German National Cohort (NAKO Gesundheitsstudie): A Deep-Learning 3T MRI Study

## TL;DR

This study uses deep learning to analyze muscle mass in over 27,000 people, finding that psoas and gluteus muscles vary with age, gender, and BMI.

## Contribution

The study introduces a high-performance deep learning model for automated 3D MRI segmentation of psoas and gluteus muscles in a large population cohort.

## Key findings

- Males had significantly higher psoas and gluteus muscle measurements compared to females.
- Younger participants (<30 years) had the highest psoas cross-sectional area, while those aged 30–59 had the highest gluteus volume.
- Higher BMI was associated with increased muscle cross-sectional area and volume in both muscles.

## Abstract

Background/Objective: This study aimed to develop and validate an automated deep learning-based model for 3D segmentation and quantification of the psoas major and gluteus muscles at 3T MRI in a large cohort study and to analyze the distribution of findings as well as gender-, age-, and BMI-related differences. Methods: The study population consisted of 27,805 participants from the MR imaging subgroup of the population-based, longitudinal German National Cohort study. A deep learning segmentation model was trained, tested, and implemented to automatically quantify psoas major maximum cross-sectional area (CSApsoas) and gluteus volume (Vgluteus) on T1-weighted 3D VIBE DIXON sequences. Associations with gender, age, and BMI were assessed by linear regression. Results: The segmentation model demonstrated a high performance, with mean Dice coefficients of 0.92 for the psoas and 0.95 for the gluteus. Males showed higher total CSApsoas (males: 37.92 ± 5.80 cm2; females: 24.47 ± 3.65 cm2) and higher total Vgluteus (males: 3.384 ± 0.528 L; females: 2.386 ± 0.408 L) compared to females. Younger participants aged <30 years showed the highest CSApsoas, whereas participants aged 30–59 years showed the highest Vgluteus. Participants with higher BMI > 25 kg/m2 showed higher muscle CSA and volumes compared to subjects with lower BMI < 25 kg/m2. Vgluteus showed a strong correlation to body weight in both females and males. Conclusions: Deep learning-based models provide accurate 3D segmentation and quantification of skeletal muscle compartments from MR images in large cohort studies, thus offering a feasible method for skeletal muscle evaluation. The morphometric size characteristics of the psoas and gluteus muscles are dependent on gender and BMI. Deep learning enables accurate 3D segmentation and quantification of skeletal muscle in large MR imaging cohorts, providing a feasible tool for muscle evaluation. The morphometric characteristics of psoas and gluteus muscles are dependent on gender and BMI.