Introduction

Sarcopenia is characterized by loss of muscle mass and function due to age. With an estimated prevalence rate of 10% in people over 60 years old, sarcopenia is associated with impaired movement, decreased quality of life, and increased morbidity and mortality. 

However, sarcopenia is not well-recognized in clinical practice, as many practitioners are unaware of diagnostic tools and instruments. While there are methods to estimate muscle mass, such as dual-energy x-ray absorptiometry (DXA) and bioelectrical impedance analysis (BIA), these methods are not always accurate or consistent. 

Given these challenges, applying deep learning algorithms to computed tomography scans presents a promising avenue for accurately determining muscle mass. Vu et al. (2024) developed a deep learning 3D segmentation algorithm that accurately determines the muscle mass of 12 abdominal muscle groups. However, this algorithm was only applied to 295 patients.

The goal of this study was to apply the deep learning algorithm to a much larger dataset and identify the trends and risk factors associated with muscle mass loss. 

Methods

The convolutional neural network developed by Vu et al. (2024) was applied to abdominal CT scans in the Penn Medicine BioBank (PMBB) to segment abdominal organ volume.

The following 12 muscle groups were segmented: left and right psoas, quadratus lumborum, erector spinae, gluteus medius, rectus abdominis, and lateral abdominals. Additionally, visceral and subcutaneous fat were segmented for each patient to understand overall body composition and fatty replacement of muscle. 

A Phenome-Wide Association Study (PheWAS) was conducted to identify the phenotypes that were statistically associated with muscle volume. The covariates were age, sex, race, and BMI, and the statistical significance threshold was calculated to be p < 2.82E-5 as determined by the Bonferroni correction. 

Results/Discussion

It was found that muscle volume consistently declined with age, beginning from ages 51-60, in both males and females. Male and female muscle volume peaked at 3.07*10^6 mm^3 and 2.14*10^6 mm^3, respectively, for patients with ages 41-50, before declining to 2.21*10^6 mm^3 for males and 1.47*10^6 mm^3 for females above age 91. 

In contrast, fat volume increased with age until it began to decline in patients with ages 71-80. In terms of the analysis of each type of fat, females had more subcutaneous fat volume than males, while males had much more visceral fat. While subcutaneous fat volume did not increase much with age, visceral fat volume dramatically increased with age, peaking at ages 71-80 for males and 61-70 for females. 

For the PheWAS, several phenotypes were found to be statistically associated with muscle volume. Notably, cachexia (wasting syndrome) and obesity were negatively associated with muscle volume. However, most of the significant phenotypes had positive associations with muscle volume, which warrants further investigation. 

Overall, the deep learning algorithm provided a quantitative understanding of body composition, including muscle and fat volume, and how it changes with age and gender. Future research could involve running Genome-Wide Association Studies (GWAS) to further understand the genetic variants and underlying risks associated with sarcopenia.