Leg Muscle Volume, Intramuscular Fat and Force Generation: Insights From a Computer-Vision Model and Fat-Water MRI

Abstract

Background: Maintaining skeletal muscle health (i.e., muscle size and quality) is crucial for preserving mobility. Decreases in lower limb muscle volume and increased intramuscular fat (IMF) are common findings in people with impaired mobility. We developed an automated method to extract markers of leg muscle health, muscle volume and IMF, from MRI. We then explored their associations with age, body mass index (BMI), sex and voluntary force generation.

Methods: We trained (n = 34) and tested (n = 16) a convolutional neural network (CNN) to segment five muscle groups in both legs from fat-water MRI to explore muscle volume and IMF. In 95 participants (70 females, 25 males, mean age [standard deviation] = 34.2 (11.2) years, age range = 18-60 years), we explored associations between the CNN measures and age, BMI and sex, and then in a subset of 75 participants, we explored associations between CNN muscle volume, CNN IMF and maximum plantarflexion force after controlling for age, BMI and sex.

Results: The CNN demonstrated high test accuracy (Sørensen-Dice index ≥ 0.87 for all muscle groups) and reliability (muscle volume ICC_2,1 ≥ 0.923 and IMF ICC₂,₁ ≥ 0.815 for all muscle groups) compared to manual segmentation. CNN muscle volume was positively associated with BMI across all muscle groups (p ≤ 0.001) but not with age (p ≥ 0.406). CNN IMF was positively associated with age for all muscle groups (p ≤ 0.015), and CNN IMF was positively associated with BMI for all muscle groups (p ≤ 0.043) except the right deep posterior compartment (p = 0.130). Males had greater CNN volume of all muscle groups (p < 0.001) except the left and right gastrocnemius (p ≥ 0.067). Gastrocnemius CNN IMF was greater in females (p ≤ 0.043). Plantarflexion force was positively associated with lateral compartment, soleus and gastrocnemius CNN volume (p ≤ 0.025) but not with CNN IMF (p ≥ 0.358).

Conclusions: Computer-vision models combined with fat-water MRI permits the non-invasive, automatic assessment of leg muscle volume and IMF. Associations with age, BMI and sex are important when interpreting these measures. Markers of leg muscle health may enhance our understanding of the relationship between muscle health, force generation and mobility.

Trial registration: ClinicalTrials.gov identifier: NCT02157038.

Keywords: computer‐assisted; image processing; leg; magnetic resonance imaging; muscle strength; rehabilitation.

FIGURE 1

Automatic segmentation of leg muscle groups. Muscle segmentations from Rater 1, Rater 2 and the convolutional neural network (CNN) from a randomly selected testing dataset are shown. (A) Muscle segmentations at the upper, middle and lower leg are overlaid a water image to show changes in muscle morphometry along the superior–inferior axis of the legs. The muscles segmented included the anterior compartment (left = dark blue, right = light blue), deep posterior compartment (left = magenta, right = orange), lateral compartment (left = green, right = purple), soleus (left = brown, right = green‐yellow) and gastrocnemius (left = gold, right = white). L = left, R = right, A = anterior, P = posterior, S = superior, I = inferior.

FIGURE 2

Relationship between convolutional neural network (CNN) muscle volume (mL), age and body mass index (BMI). Partial correlations (Pearson's r) were performed to identify linear relationships between CNN volume and age or CNN volume and BMI in 95 participants (70 females, 25 males, age = 34.2 [11.2] years, body mass index [BMI] = 25.1 [4.5] kg/m²) after controlling for sex and BMI or sex and age, respectively. For CNN volume and age, the residuals of volume are plotted on the residuals of age after controlling for sex and BMI. For CNN volume and BMI, the residuals of volume are plotted on the residuals of BMI after controlling for sex and age. CNN muscle volume was not associated with age for all muscle groups. CNN muscle volume was positively associated with BMI for all muscle groups. See Table 2 for the results from multiple linear regression analysis with factors of age, BMI and sex. AC = anterior compartment, DPC = deep posterior compartment, LC = lateral compartment, Gastroc = gastrocnemius.

FIGURE 3

Relationship between convolutional neural network (CNN) intramuscular fat (IMF, %), age and body mass index (BMI). Partial correlations (Pearson's r) were performed to identify linear relationships between CNN IMF and age or CNN IMF and BMI in 95 participants (70 females, 25 males, age = 34.2 [11.2] years, BMI = 25.1 [4.5] kg/m²). For CNN IMF and age, the residuals of IMF are plotted on the residuals of age after controlling for sex and BMI. For CNN IMF and BMI, the residuals of IMF are plotted on the residuals of BMI after controlling for sex and age. CNN IMF was positively associated with age for all muscle groups. CNN IMF was positively associated with BMI for all muscle groups except the right deep posterior compartment. See Table 2 for the results from multiple linear regression analysis with factors of age, BMI and sex. AC = anterior compartment, DPC = deep posterior compartment, LC = lateral compartment, Gastroc = gastrocnemius.

FIGURE 4

Sex differences in convolutional neural network (CNN) muscle volume (mL) and CNN intramuscular fat (IMF, %) in 95 participants (70 females, 25 males, age = 34.2 (11.2) years, body mass index (BMI) = 25.1 [4.5] kg/m²). (A) CNN muscle volume by sex for each muscle. Males had sign larger CNN muscle volume than females for the left and right anterior compartment, deep posterior compartment, lateral compartment, and soleus but not the left and right gastrocnemius (controlling for age and BMI). (B) CNN IMF by sex for each muscle. Females had higher CNN IMF than males for the left and right gastrocnemius (controlling for age and BMI). Estimated marginal means are shown. See Table 2 for results from multiple linear regression with factors of age, BMI and sex. Error bars = 1 standard error. *p < 0.05, ***p < 0.001. AC = anterior compartment, DPC = deep posterior compartment, LC = lateral compartment, Gastroc = gastrocnemius.