# Automated Segmentation of Forearm Muscles: Clinical Associations With Hand Function, Muscle Volume and Intramuscular Fat

## TL;DR

This study develops an automated method to analyze forearm muscle health using MRI scans and finds that muscle volume is linked to grip strength and body mass index.

## Contribution

A computer-vision model for automated forearm muscle segmentation is developed and validated, with clinical associations explored.

## Key findings

- The computer-vision model achieved high accuracy in segmenting forearm muscles with Sørensen–Dice indices of 0.89 and 0.85 for flexors and extensors.
- Muscle volume was positively correlated with BMI and grip strength, but not with age or intramuscular fat.
- Males had larger muscle volumes than females, but no sex differences were found in intramuscular fat.

## Abstract

Hand function is critical for daily activities and declines early in many diseases, conditions or disorders affecting the musculoskeletal and neurologic systems. Muscle health markers derived from clinically available magnetic resonance imaging (MRI) scans are strongly associated with functional capacity, may enhance clinical assessment and inform management options. However, traditional muscle MRI assessments require time‐intensive manual segmentations. Here, we aim to develop and test a computer‐vision model for automated forearm muscle segmentation and investigate associations between MRI‐derived muscle markers and age, sex, BMI, functional grip strength and dexterity measures.

We recruited 42 healthy, right‐handed adults (54.8% female, median age 37.3 years, median BMI: 23.0). Grip strength and dexterity were measured using the NIH Toolbox motor battery. Dixon fat‐water MRI of the right forearm was acquired at 3.0 T, and forearm flexor and extensor muscle compartments were manually segmented for model training. A 2D U‐Net convolutional neural network model was trained and tested for segmentation of the forearm flexors and extensors for the assessment of muscle volume and intramuscular fat. Testing accuracy and reliability were assessed using Sørensen–Dice indices, intraclass correlation coefficients (ICCs) and Bland–Altman analyses. Associations between the MRI‐derived muscle markers, demographic factors, muscle metrics and hand function were evaluated using partial correlations and regression models.

The segmentation model showed high test accuracy, achieving mean Sørensen–Dice indices of 0.89 (flexors) and 0.85 (extensors) and ICCs of 0.75–0.99 for muscle volume and intramuscular fat. Muscle volume was positively correlated with BMI (p < 0.001) but not age (p > 0.249). Males had larger muscle volumes than females (p < 0.001), with no sex differences in intramuscular fat (p > 0.141), and no association between intramuscular fat and grip strength or dexterity (p > 0.350). We observed strong positive correlations between grip strength and both flexor (p = 0.004) and extensor (p = 0.001) muscle volumes, while dexterity showed no significant associations.

Our findings highlight the accuracy and reliability of automated forearm muscle segmentation using computer vision. BMI emerged as a key determinant of muscle volume, independent of age. The strong association between muscle volume and grip strength demonstrates the clinical relevance of these metrics, suggesting potential applications in therapeutic planning for conditions impairing hand function. Sex‐based differences in muscle volume underscore the importance of tailored assessments. Computer vision models integrated with Dixon fat‐water MRI enable efficient, accurate evaluation of forearm muscle health. Future research should explore these metrics in clinical populations and their utility in tracking functional outcomes.