Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Mar 6.
Published in final edited form as: Obesity (Silver Spring). 2025 Mar 6;33(Suppl 1):96–109. doi: 10.1002/oby.24243

Musculoskeletal Characteristics in Older Adults with Overweight or Obesity: INVEST in Bone Health Trial Baseline Analysis

S Delanie Lynch 1, Marjorie Howard 2, Daniel P Beavers 3, Leon Lenchik 4, Ryan Barnard 2, Joshua R Stapleton 1, Erica Lawrence 5, Peggy M Cawthon 6, Fang-Chi Hsu 2, Kristen M Beavers 7, Ashley A Weaver 1
PMCID: PMC12353373  NIHMSID: NIHMS2050364  PMID: 40051020

Abstract

Objective:

Examine associations of computed tomography (CT)-derived musculoskeletal measures with demographics and traditional musculoskeletal characteristics.

Methods:

The INVEST in Bone Health trial (NCT04076618) acquired a battery of musculoskeletal measures in 150 older adults living with overweight or obesity. At baseline, CT (volumetric bone mineral density, cortical thickness, muscle radiomics, muscle/intermuscular adipose tissue [IMAT] area and density), dual-energy x-ray absorptiometry [DXA] (areal bone mineral density, total body fat mass, appendicular lean mass, lean body mass) and strength assessments (grip strength, knee extensor strength) were collected, along with demographic and clinical characteristics. Analyses employed linear regression and mixed effects models along with factor analysis for dimensionality reduction of the radiomics data.

Results:

Participants were older (66±5yrs), mostly female (75%) and living with overweight or obesity (body mass index [BMI]:33.6±3.3kg/m2). Age was not significantly associated with most CT-derived bone, IMAT or muscle measures. BMI was significantly associated with DXA and CT-derived muscle and IMAT measures, which were higher in males than females (all p<0.01). For the mid-thigh, muscle size was significantly related to grip and knee extensor strength (both p<0.01).

Conclusions:

Machine learning-derived CT metrics correlated strongly with DXA and muscle strength, with higher BMI linked to greater IMAT and poorer muscle quality.

Keywords: computed tomography, age, obesity, muscle, radiomics

Introduction

By 2060, projections indicate that adults aged 65 and older will represent 24% of the US population, a nearly 10% increase from 20191. Among this demographic, preserving musculoskeletal health is crucial due to age-related physiological declines, including bone and muscle mass loss2. Concurrently, obesity rates among this population have surged, reaching 41.5% in the last decade3. This dual challenge of aging and obesity accelerates musculoskeletal health decline, giving rise to conditions like sarcopenic or osteosarcopenic obesity4. With both these populations increasing, it is a priority to understand the effect of these factors to inform interventions and mitigate musculoskeletal health burden.

Although widely utilized and economically accessible, dual-energy x-ray absorptiometry (DXA) suffers from notable limitations in the clinical and research arenas5, 6. These limitations include an inability to capture compartmental properties of bone which are crucial for assessing fracture risk, particularly in populations living with obesity. Further, excessive adipose tissue associated with obesity has been shown to decrease the accuracy of DXA bone measures7. With regard to muscle, DXA is only able to approximate lean mass as a surrogate measure for muscle size and this approximation is inconsistently associated with muscle strength8. These limitations pose a significant challenge in characterizing musculoskeletal health for a growing segment of the population.

To address these shortcomings, computed tomography (CT) emerges as a promising alternative. Recent advancements, particularly in machine learning for musculoskeletal segmentation, have enhanced the accuracy and accessibility of CT imaging. Although associated with higher cost and radiation than DXA, CT offers superior sensitivity in detecting fracture risk by providing separate bone mineral density (BMD) measures of trabecular and cortical bone compartments9–11. Moreover, it enables automatic acquisition of precise muscle health metrics, including area and density, allowing for assessments of muscle quantity and quality with better reliability compared to other modalities12. Despite their potential clinical promise, few studies have investigated the associations between aging, obesity, and CT musculoskeletal measurements. Additionally, correlations between CT muscle measures and traditional metrics, such as DXA lean mass and muscle strength, remain largely unexplored, particularly for the intersection of obesity and aging.

The INVEST in Bone Health clinical trial presents an ideal opportunity to address these knowledge gaps. Here, we aim to: (1)Examine associations between age, body mass index (BMI) and sex with CT-derived measures of bone, muscle and intermuscular adipose tissue (IMAT), (2)Determine associations between CT-derived muscle and IMAT metrics (area and density) with total body weight and traditional metrics (DXA lean and fat masses, and muscle strength), (3)Evaluate correlations between CT-derived (volumetric BMD) and traditional DXA-derived (areal BMD) measures of bone, (4)Investigate cross-sectional correlations of novel CT-derived muscle radiomics in this population with age, BMI, CT-derived muscle measurements and traditional metrics (DXA lean and fat masses, and muscle strength).

Methods

Overview of the INVEST in Bone Health clinical trial

This is a cross-sectional analysis of baseline data collected from September 2019 through April 2023 for older adults enrolled in the INVEST in Bone Health clinical trial (R01AG059186). Individuals were eligible for the study if they were 60-85 years old, living with an indication for weight loss (BMI 30-40 kg/m2 or 27.0≤30.0 kg/m2 plus one obesity-related risk factor/clinical comorbidity), had a stable body weight (no weight loss >5% in past 6-months), and were willing/able to participate in all study procedures/assessments. The trial is registered on ClinicalTrials.gov (NCT04076618) and was approved by the Wake Forest University School of Medicine Institutional Review Board (IRB No.00058279). Full trial design details have been published13. Below we highlight baseline assessments and methods relevant to this analysis.

Sample demographic and clinical characteristics

Age, race, ethnicity, educational and medical history were self-reported by participants. Weight was measured to the nearest 0.1 kg using a balance beam scale and standing height was measured to the nearest 0.1 cm (Tanita #WB-3000PLUS). Measured weight and height were used to calculate BMI (kg/m2). Additionally, the Community Healthy Activities Model Program for Seniors (CHAMPS14) physical activity questionnaire was administered to quantify baseline levels of caloric expenditure per week for all activities and moderate-intensity activities.

Muscle strength

As previously described15, grip strength (kg) was measured twice in each hand with a Jamar hydraulic hand dynamometer (Performance Health, Warrenville, IL) and mean value of the dominant hand was included in analysis. Knee extensor strength, as peak torque (Nm), was measured via isokinetic dynamometry (Biodex, Shirley, NY) in the dominant leg16.

Dual-energy X-ray absorptiometry (DXA)

All participants underwent whole-body, hip, and spine DXA scanning with a GE iDXA (Medical Systems, Madison, WI). From these scans, areal BMD (aBMD) was derived for the total hip, femoral neck, and lumbar spine. Additional parameters for soft tissue were acquired with DXA including total body fat mass (kg), total body lean mass (kg) and appendicular lean mass (kg), which encompasses the bone-free lean body mass of the upper and lower extremities. All scans were performed in accordance with manufacturer recommended positioning and reviewed by an International Society of Clinical Densitometry certified DXA technologist.

Computed tomography (CT)

Figure 1 provides an overview of the CT methods described below.

Figure 1.

Figure 1.

Open in a new tab

Image processing methods for all CT musculoskeletal measurements.

Scan acquisition

A helical CT scan was obtained on a 64-slice PET/CT GE Discovery MI scanner with standard abdomen reconstruction filters at 2.5-mm slice thickness. Scans covered from mid-chest to ~3-cm below mid-shaft of the femurs. Scans were acquired with 50-cm field of view, 120 kV, 1:1 pitch, and auto-exposure. A 5-port bone mineral phantom and bolus bag (Mindways Software, Inc., Austin, TX) was positioned under each participant, covering the lumbar spine through the lesser trochanter. Monthly calibration was performed in accordance with the manufacturer’s instructions by trained technologists.

CT-derived bone measurements

Using QCTPro™ software (Mindways, Austin, TX), volumetric BMD (vBMD) was obtained for the lumbar vertebrae (L1-L4) and proximal femurs (bilaterally). The phantom imaged in each scans was used to calibrate vBMD (mg/cm3) to equivalent aqueous K2HPO4 density17. The proximal femur was automatically divided to obtain trabecular and cortical vBMD measures for the total hip, femoral neck, intertrochanteric, and trochanteric regions, as well as calculate integral vBMD by combining both compartments. For the spine, an elliptical region of interest was placed in the center of each vertebra to obtain trabecular vBMD.

For segmentation-based analyses, the dominant femur and L3 vertebrae were chosen as representative regions to minimize computational effort. Femurs were semi-automatically segmented using thresholding in Mimics (v24, Materialise) and L3 vertebrae were automatically segmented using Anduin Bonescreen18. Segmentations were finalized and reviewed by an experienced reader. Resulting surface models were used to measure cortical thickness distributions with a cortical density-based algorithm implemented in Stradview (v7.2, Cambridge University, UK)19. Average cortical thickness over each surface was calculated.

CT-derived muscle and IMAT measurements

From each CT, single slice DICOMs were manually selected at the mid-L3 vertebral and mid-thigh levels. Mid-thigh was defined as half the length between the superior aspect of the greater trochanter and the inferior aspect of the lateral condyle. These DICOMs were input into our group’s machine learning algorithms, which adopt different convolutional neural network architectures, including U-Net and DeepLabV3+U-Net20–22, to automatically segment muscle and adipose tissue. Separate models have been trained for each region and are configured on our institution’s high performance computing cluster23. Manual corrections on these segmentations were performed as needed. All skeletal muscle and IMAT areas (cm2) and densities (HU) were extracted from these segmentations. Muscle area is indicative of quantity, where higher area correlates to more muscle mass. Muscle density is inversely related to IMAT, with increasing density correlating to less fat infiltration in muscle and better muscle quality.

The open-source platform, PyRadiomics24, was used to derive radiomic features of muscle using L3 muscle segmentations12. This platform automatically extracts multi-dimensional measures via statistical modeling of voxel distribution within an image to characterize heterogeneity of muscle tissue. A total of 75 features were obtained across 5 radiomic classes:

  1. Gray-level size zone matrix (GLSZM;16 features): Quantifies an image’s gray level zones which are the number of connected voxels that share the same gray level intensity.

  2. Gray-level co-occurrence matrix (GLCM;24 features): Characterizes texture by measuring how often pixel pairs of unique values with certain spatial relationships occur in an image.

  3. Gray-level run length matrix (GLRLM;16 features): Quantifies gray level runs, defined as the length of consecutive pixels with the same gray level value.

  4. Gray-level dependence matrix (GLDM;14 features) Characterizes gray level dependencies, defined as the number of connected voxels within a given distance that are dependent on the center voxel.

  5. Neighboring gray-tone difference matrix (NGTDM;5 features): Measures the difference between a gray value and the average gray value of its neighbors within a certain distance.

Statistical analyses

Baseline characteristics were summarized using descriptive statistics. Linear regression and mixed effects models were used to explore associations between age, BMI and sex with CT-derived bone, muscle and IMAT measures. Mixed effects models with a subject-specific random intercept and unstructured covariance were used for hip/thigh CT metrics to account for bilateral measurements. Linear regression models, adjusted for age and sex only, were used to examine associations between CT-derived muscle and IMAT metrics (area and density) with outcome variables including total body weight and traditional metrics (DXA lean and fat masses, and muscle strength). For metrics collected on both the left and right sides (e.g., mid-thigh CT muscle area), the left side was used for associations with DXA variables, while the dominant side was used for associations with muscle strength variables. To determine associations between vBMD and aBMD measures derived from CT and DXA, Spearman’s rank correlations were used with adjustments for age, sex and BMI. DXA variables were acquired only on the left side and therefore, this correlation was similarly performed with the left side only for bilateral CT measures.

Radiomic features were standardized25 and factor analysis26 was employed for dimensionality reduction to define a smaller set of variables explaining the underlying data. In our analysis, varimax rotation was applied to factor loadings. Spearman’s rank correlation was then used to explore associations between the resulting factors with age, BMI, CT-derived muscle measurements and traditional metrics (DXA lean and fat masses, muscle strength).

Analyses were performed using SAS v. 9.4 (SAS Institute Inc. Cary, NC) and R software, with p-values of 0.05 used to determine statistical significance. Due to the exploratory nature of this work, analyses were not adjusted for multiple comparisons.

Results

Sample demographic and clinical characteristics

Baseline descriptive characteristics of the 150 older adults (66.4±4.6 years old) enrolled in the INVEST in Bone Health trial are summarized in Table 1. Most participants were Non-Hispanic (98%), White/Caucasian, (69%) females (75%). At the time of randomization, average BMI was 33.9±3.2 kg/m2 for men and 33.5±3.4 kg/m2 for women. Overall, 14.7% of our population was defined as overweight (25≤BMI<30 kg/m2) and 85.3% were defined as obese (BMI≥30 kg/m2). Self-reported physical activity expenditure was 2527.8±1862.5 and 672.8±1110.4 calories per week for all activities and moderate-intensity activities, respectively.

Table 1.

Baseline descriptive characteristics of randomized participants overall and stratified by sex, presented as N (%) or mean ± SD.

Overall (N=150) Males (N=38) Females (N=112)
Age, years 66.4 ± 4.6 68.1 ± 4.2 65.8 ± 4.6
Race
  White/Caucasian 103 (68.7) 30 (78.9) 73 (65.2)
  Black/African American 43 (28.7) 7 (18.4) 36 (32.1)
  Multi-racial or Other 4 (2.7) 1 (2.6) 3 (2.7)
Ethnicity
  Hispanic/Latino 3 (2) 0 (0) 3 (2.7)
  Non-Hispanic/Latino 147 (98) 38 (100) 109 (97.3)
Body Mass (kg) 92.9 ± 12.8 105.1 ± 11 88.7 ± 10.5
Height (cm) 166.2 ± 8.4 176.2 ± 6 162.7 ± 6
Body Mass Index (kg/m2) 33.6 ± 3.3 33.9 ± 3.2 33.5 ± 3.4
  Overweight (25 ≤ BMI < 30 kg/m2) 22 (14.7) 4 (10.5) 18 (16.1)
  Obese (BMI ≥ 30 kg/m2) 128 (85.3) 34 (89.5) 94 (83.9)
Education
  High School/Equivalent (09-12) 28 (18.7) 9 (23.7) 19 (17)
  College (13-16) 81 (54) 17 (44.7) 64 (57.1)
  Post-Graduate 41 (27.3) 12 (31.6) 29 (25.9)
Diabetes 12 (8) 4 (10.5) 8 (7.1)
Pre-existing low bone mass 1 73 (48.7) 15 (39.5) 58 (51.8)
CHAMPS Physical Activity (caloric expenditure/week: all activities) 2 2527.8 ± 1862.5 (2029.3 [1179.4, 3262.9]) 3055.9 ± 2210 (2520.7 [1403.3, 3944.9]) 2348.7 ± 1703.1 (1934.3 [1132.3, 2992.0])
CHAMPS Physical Activity (caloric expenditure/week: moderate-intensity activities) 2 672.8 ± 1110.4 (204.6 [0.0, 791.4]) 1127.1 ± 1435.2 (602.7 [0.0, 1719.9]) 518.7 ± 934.8 (0.0 [0.0, 663.6])
Open in a new tab
1

Pre-existing low bone mass: t-score on hip, spine, forearm radius, or femoral neck DXA of −1 to −2.49

2

Due to skewness, (Median [IQR]) are provided for these variables in addition to mean ± SD

Associations between age, BMI and sex with CT-derived measures of bone, muscle and IMAT

Table 2 highlights adjusted associations between the baseline CT musculoskeletal metrics (unadjusted values in Table S1) with age, BMI, and sex. Notably, age was not significantly associated with the majority of CT bone, muscle or IMAT measures except for L3 muscle density and intertrochanteric cortical vBMD. After adjusting for sex and BMI, a one-year increment in age was associated with lower L3 muscle density by 0.29 HU (p=0.002) and higher intertrochanteric cortical vBMD by 1.29 mg/cm3 (p=0.041).

Table 2.

Associations between age, BMI, and sex with baseline CT-derived measures of bone, muscle and IMAT, presented by region with significant values, defined as p<0.05, in bold.

Age BMI Sex
Slope (95% CI) p-value Slope (95% CI) p-value Male Mean ± SE Female Mean ± SE Mean (95% CI) Difference Male-Female p-value
Region Muscle/Fat Variables
L3 Level2 Muscle Area (cm2) −0.76 (−1.73, 0.21) 0.122 2.53 (1.22, 3.84) < .001 178.45 ± 4.44 124.41 ± 2.56 54.03 (43.81, 64.26) < .001
Muscle Density (HU) −0.29 (−0.48, −0.11) 0.002 −0.21 (−0.46, 0.04) 0.106 36.96 ± 0.86 31.39 ± 0.49 5.58 (3.6, 7.55) < .001
IMAT Area (cm2) 0.24 (0, 0.49) 0.052 0.45 (0.12, 0.78) 0.009 15.09 ± 1.13 13.91 ± 0.65 1.17 (−1.42, 3.77) 0.372
Mid-Thigh Level1 Muscle Area (cm2) −0.24 (−0.8, 0.31) 0.387 1.5 (0.72, 2.28) < .001 160.71 ± 2.7 114.69 ± 1.46 46.02 (39.92, 52.12) < .001
Muscle Density (HU) −0.08 (−0.22, 0.05) 0.237 −0.22 (−0.42, −0.03) 0.023 44.53 ± 0.66 44.79 ± 0.36 −0.26 (−1.76, 1.25) 0.737
IMAT Area (cm2) −0.07 (−0.18, 0.05) 0.251 0.56 (0.4, 0.72) < .001 11.84 ± 0.55 9.47 ± 0.3 2.37 (1.13, 3.61) < .001
Bone Variables
Total Hip1 Cortical vBMD (mg/cm3) 0.94 (−0.01, 1.9) 0.053 −1.04 (−2.33, 0.25) 0.112 709.83 ± 4.36 698.91 ± 2.53 10.92 (0.86, 20.97) 0.034
Trabecular vBMD (mg/cm3) 0.15 (−0.62, 0.92) 0.700 0.61 (−0.43, 1.64) 0.247 129.44 ± 3.5 132.79 ± 2.03 −3.35 (−11.42, 4.72) 0.413
Femoral Neck1 Cortical vBMD (mg/cm3) 0.42 (−0.84, 1.68) 0.509 −2.19 (−3.88, −0.49) 0.012 686.32 ± 5.73 697.57 ± 3.33 −11.25 (−24.48, 1.97) 0.095
Trabecular vBMD (mg/cm3) 0.13 (−0.77, 1.04) 0.771 0.83 (−0.38, 2.05) 0.178 125.21 ± 4.12 128.19 ± 2.39 −2.98 (−12.47, 6.52) 0.536
Trochanter1 Cortical vBMD (mg/cm3) 0.08 (−1.11, 1.28) 0.889 −0.93 (−2.53, 0.68) 0.255 545.9 ± 5.43 539.05 ± 3.15 6.85 (−5.68, 19.37) 0.282
Trabecular vBMD (mg/cm3) 0.43 (−0.24, 1.11) 0.204 0.41 (−0.49, 1.31) 0.371 129.61 ± 3.06 133.09 ± 1.78 −3.48 (−10.54, 3.57) 0.330
Intertrochanteric1 Cortical vBMD (mg/cm3) 1.29 (0.06, 2.52) 0.041 −0.85 (−2.51, 0.82) 0.316 771.64 ± 5.62 754.6 ± 3.26 17.04 (4.07, 30) 0.010
Trabecular vBMD (mg/cm3) −0.03 (−0.98, 0.91) 0.944 0.73 (−0.53, 2) 0.255 130.44 ± 4.29 133.7 ± 2.49 −3.26 (−13.15, 6.63) 0.516
Lumbar Spine2 L1-L4 Average vBMD (mg/cm3) 0.15 (−1.18, 1.49) 0.820 −0.04 (−1.85, 1.76) 0.961 117.9 ± 6.11 121.75 ± 3.53 −3.85 (−17.92, 10.22) 0.590
L3 Cortical Thickness (mm) 0.007 (−0.006, 0.02) 0.308 0.007 (−0.011, 0.025) 0.428 3.02 ± 0.06 2.77 ± 0.04 0.253 (0.112, 0.394) < .001
Dominant Femur2 Cortical Thickness (mm) 0.004 (−0.005, 0.013) 0.329 0.012 (0, 0.024) 0.053 1.83 ± 0.04 1.69 ± 0.02 0.145 (0.05, 0.24) 0.003
Open in a new tab

Results are presented as slope and 95% confidence interval or adjusted mean (least squares mean) ± standard error with p-values.

1

Hip/Thigh CT Measurements: Mixed effects models, with random intercept and unstructured covariance to account for measurements on right and left sides, are adjusted for age, BMI, and sex.

2

Other CT Measurements: Linear regression models are adjusted for age, BMI, and sex.

Several notable connections were found between BMI and muscle as well as IMAT areas, while only one bone variable showed a significant correlation with BMI. For each unit increase in BMI (+1 kg/m2), area was higher by 2.53 cm2 for L3 muscle (p<0.001), and 1.50 cm2 for mid-thigh muscle (p<0.001). Additionally, each unit increment in BMI was associated with higher IMAT areas for both L3 (0.45 cm2; p<0.001) and mid-thigh (0.56 cm2; p<0.001) regions. BMI was not strongly associated with regional muscle densities, cortical thicknesses, or vBMD measures except for higher BMI (+1 kg/m2), being associated with lower femoral neck cortical vBMD (2.19 mg/cm3; p=0.012).

Muscle area was significantly higher in males compared to females (both p<0.001) at each anatomic location. Sex-based differences, adjusted for age and BMI, were most notable for L3 muscle area, with a male-female difference of 54.03 cm². Males exhibited higher L3 muscle density, by a factor of 5.58 HUs, compared to females (p<0.001), but there was no statistically significant difference by sex for the mid-thigh. Similarly, cortical thicknesses of the spine (p<0.001) and femur (p=0.003) were higher in males compared to females. Although most vBMD measures were not significantly different between the two sexes, we did observe that males had higher cortical vBMD for the total hip (10.92 mg/cm3; p=0.034) and intertrochanteric regions (17.04 mg/cm3; p=0.010).

Associations between CT-derived metrics and traditional metrics

Table 3 presents associations of CT-derived muscle quantity and quality measures, excluding radiomics, with traditional metrics including DXA (total body fat mass, lean body mass, appendicular lean mass) and muscle strength (grip and knee extensor strength). The linear models used for these associations include covariate adjustments to minimize confounding effects of age and sex and utilize the measurement from the left side or dominant side when applicable. Positive linear relationships were observed between body mass and CT muscle (L3 slope: +0.1, p=0.002; Mid-thigh slope: +0.26, p<0.001) and IMAT areas (L3 slope: +0.45, p<0.001; Mid-thigh slope: +1.19, p<0.001) whereas negative linear relationships were found between body mass and CT muscle densities (L3 slope: −0.38, p=0.025; Mid-thigh slope: −0.41, p=0.06). DXA total body fat mass was significantly associated with CT IMAT areas (both p<0.001), with each unit increment in L3 and mid-thigh IMAT areas (+1 cm2) associated with higher fat masses of 0.31 kg and 0.85 kg, respectively. DXA lean body mass and appendicular lean mass were both significantly associated with CT muscle areas (all p<0.001) but not CT muscle densities (all p>0.521). Incremental increases in L3 and mid-thigh muscle areas (+1 cm2) were associated with greater grip strengths of 0.08 kg and 0.16 kg, respectively (both p<0.001). For mid-thigh muscle, higher muscle area (+1 cm2) and density (+1 HU) were each strongly associated with greater knee extensor strength (0.73 Nm and 2.09 Nm; both p<0.001).

Table 3.

Associations between regional CT-derived muscle and IMAT variables and total body mass, DXA-derived lean and fat mass variables, and muscle strength measures. Significant associations, defined as p<0.05, are shown in bold.

graphic file with name nihms-2050364-t0003.jpg
Open in a new tab

Linear regression models include adjustments for age and sex.

1

Indicates associations where left side was used for mid-thigh CT measures as the corresponding side for available DXA measurements

2

Indicates associations where dominant side was used for mid-thigh CT measures as the corresponding side for available muscle strength measurements

Correlations between CT-derived vBMD and DXA-derived aBMD

Adjusted correlations between CT integral vBMD and DXA aBMD for the total hip, femoral neck and lumbar spine regions are presented in Table 4. For each region and across different regions, vBMD and aBMD were significantly correlated (all p<0.001), with the strongest correlation between total hip vBMD and total hip aBMD (r=0.90) and the weakest correlation between femoral neck vBMD and lumbar spine aBMD (r=0.54).

Table 4.

Correlations between regional, volumetric bone mineral densities [vBMD] (measured by CT) and areal bone mineral densities [aBMD] (measured by DXA). Significant associations, defined as p<0.05, are shown in bold.

graphic file with name nihms-2050364-t0004.jpg
Open in a new tab

Spearman rank correlation coefficients and p-values

*

Adjusted for age, BMI, and sex

Novel CT-derived muscle radiomics

From the 75 radiomic features, six radiomic factors were identified to explain most of the variation and represent patterns within the data. Factor loadings are provided in Table 5. Radiomic Factor 1 accounts for 37% of the variation in the radiomics data. Out of the 75 radiomic features, 37 features shown in Figure 2 can be used to interpret Radiomic Factor 1 (15 GLCM, 8 GLRLM, 7 GLDM, 6 GLSZM, 1 NGTDM). Factor Loadings for Radiomics Factors 2-6 are provided in Figures S1–S5, with each factor interpreted by the following features: Radiomic Factor 2 (18 features: 4 GLCM, 4 GLDM, 5 GLRLM, 5 GLSZM), Radiomic Factor 3 (8 features: 2 GLRLM, 2 GLDM, 2 GLSZM, 2 NGDM), Radiomic Factor 4 (6 features: 1 GLRLM, 1 GLDM, 4 GLCM), Radiomic Factor 5 (4 features; 1 GLCM, 3 GLSZM), and Radiomic Factor 6 (2 features; 2 NGTDM).

Table 5.

Radiomic factor loadings for all 75 radiomic variables, where the loading indicates both strength and direction of the relationship. Each radiomic factor can be interpreted based on the shaded variables, which are described in detail in the PyRadiomics documentation (https://pyradiomics.readthedocs.io/en/latest/features.html)

Muscle Radiomic Variables Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6
glcm_JointEntropy 0.9798 0.0355 0.1712 0.0435 −0.0515 0.0031
glrlm_RunPercentage 0.9647 −0.1032 0.1302 −0.1887 −0.0037 −0.0013
glrlm_ShortRunEmphasis 0.9621 −0.1088 0.1097 −0.1975 0.0339 0.0115
glrlm_RunLengthNonUniformityNormalized 0.9612 −0.1033 0.1157 −0.1985 0.0587 0.0098
glcm_SumSquares 0.9536 −0.0469 0.1705 0.2144 0.0509 0.0118
gldm_GrayLevelVariance 0.9483 0.0425 0.1283 0.2574 0.0641 0.0229
gldm_DependenceNonUniformityNormalized 0.9425 −0.0497 0.2121 −0.1486 −0.0094 −0.0266
glszm_ZonePercentage 0.9421 −0.1515 0.0774 −0.2005 0.1720 0.0175
gldm_SmallDependenceEmphasis 0.9342 −0.1439 0.0732 −0.2158 0.2185 0.0141
glcm_DifferenceAverage 0.9331 0.1605 0.2095 −0.2113 0.0746 −0.0158
glcm_DifferenceEntropy 0.9302 0.1639 0.2004 −0.2198 0.0846 −0.0027
glrlm_GrayLevelVariance 0.9163 0.1930 0.1449 0.2776 0.1064 0.0430
glcm_Contrast 0.9142 0.1697 0.2120 −0.2000 0.1645 −0.0051
glcm_SumEntropy 0.9063 −0.1385 0.1079 0.3523 −0.1262 0.0206
glcm_DifferenceVariance 0.9010 0.1740 0.2045 −0.2009 0.2083 0.0122
glcm_ClusterTendency 0.8725 −0.1647 0.1285 0.4263 −0.0191 0.0201
glcm_ClusterProminence 0.8331 −0.0360 0.2014 0.4241 0.1024 0.1198
ngtdm_Contrast 0.7947 0.0473 0.1082 −0.0418 0.0939 −0.5761
gldm_SmallDependenceHighGrayLevelEmphasis 0.7454 0.5613 0.2503 −0.1552 0.1778 0.0104
glszm_GrayLevelNonUniformityNormalized −0.6777 −0.5523 −0.3193 −0.1749 0.0307 0.0133
glcm_Idmn −0.7238 −0.0922 −0.1551 0.2488 −0.0795 0.6029
glcm_Idn −0.7875 −0.0983 −0.1661 0.2559 −0.0175 0.5219
glrlm_LongRunHighGrayLevelEmphasis −0.7904 0.5769 −0.0186 0.1132 0.1177 −0.0020
gldm_LargeDependenceHighGrayLevelEmphasis −0.8140 0.5461 −0.0531 0.1211 0.0855 −0.0033
glszm_LargeAreaHighGrayLevelEmphasis −0.8174 0.2895 0.0490 0.1040 0.2714 −0.0164
glszm_ZoneVariance −0.8499 0.2214 0.0314 0.0802 0.2712 −0.0178
glszm_LargeAreaEmphasis −0.8541 0.2207 0.0299 0.0832 0.2666 −0.0181
glszm_LargeAreaLowGrayLevelEmphasis −0.8726 0.0837 0.0178 0.0892 0.2634 −0.0156
glcm_MaximumProbability −0.9184 0.0803 −0.2325 −0.0681 0.0384 0.0191
glrlm_GrayLevelNonUniformityNormalized −0.9239 −0.1664 −0.0529 −0.2561 0.1055 0.0169
gldm_DependenceVariance −0.9316 0.0424 −0.2084 0.1572 0.1218 0.0427
glcm_Idm −0.9392 −0.1511 −0.2051 0.2172 −0.0090 0.0228
glcm_Id −0.9395 −0.1495 −0.2045 0.2182 0.0037 0.0241
glrlm_RunVariance −0.9527 0.1092 −0.1293 0.1701 0.1302 0.0147
glrlm_LongRunEmphasis −0.9582 0.1107 −0.1223 0.1764 0.0999 0.0071
gldm_LargeDependenceEmphasis −0.9627 0.0958 −0.1383 0.1838 0.0540 0.0070
glcm_JointEnergy −0.9638 0.0973 −0.1198 −0.0597 0.1637 0.0176
glrlm_HighGrayLevelRunEmphasis −0.0950 0.9698 0.1884 −0.0180 0.0382 −0.0088
gldm_HighGrayLevelEmphasis −0.1772 0.9641 0.1484 −0.0048 0.0503 −0.0108
glcm_SumAverage −0.2291 0.9599 0.0974 −0.0015 0.0446 −0.0184
glcm_JointAverage −0.2291 0.9599 0.0974 −0.0015 0.0446 −0.0184
glcm_Autocorrelation −0.2217 0.9553 0.1157 0.0257 0.0536 −0.0109
glrlm_ShortRunHighGrayLevelEmphasis 0.2451 0.9253 0.2465 −0.0741 0.0357 −0.0031
glszm_HighGrayLevelZoneEmphasis 0.2052 0.9034 0.3228 −0.0381 −0.0084 0.0097
glszm_SmallAreaHighGrayLevelEmphasis 0.2975 0.8493 0.3126 −0.0775 0.1558 0.0203
glszm_GrayLevelVariance 0.5367 0.6726 0.3027 0.2563 0.0812 0.0942
glcm_ClusterShade 0.6221 −0.6320 0.1329 −0.0161 0.0523 0.1651
gldm_SmallDependenceLowGrayLevelEmphasis 0.5172 −0.8195 −0.0873 −0.0123 0.1773 0.0454
gldm_LargeDependenceLowGrayLevelEmphasis −0.2941 −0.8960 −0.0852 0.2255 0.0184 0.0384
glrlm_LongRunLowGrayLevelEmphasis −0.2563 −0.9166 −0.1180 0.2193 0.0664 0.0427
gldm_LowGrayLevelEmphasis 0.3488 −0.9235 −0.0542 0.07172 0.0584 0.0377
glrlm_ShortRunLowGrayLevelEmphasis 0.3326 −0.9239 −0.0983 0.0735 0.0902 0.0436
glszm_SmallAreaLowGrayLevelEmphasis −0.0369 −0.9319 −0.2055 0.1551 0.1649 0.0416
glrlm_LowGrayLevelRunEmphasis 0.2635 −0.9461 −0.0946 0.0987 0.0720 0.0420
glszm_LowGrayLevelZoneEmphasis −0.0285 −0.9502 −0.1871 0.1790 0.0896 0.0395
glrlm_GrayLevelNonUniformity 0.1420 0.3718 0.8960 −0.0678 −0.1418 0.0158
gldm_GrayLevelNonUniformity −0.1399 0.4521 0.8627 −0.0473 −0.1105 0.0173
gldm_DependenceNonUniformity 0.3923 0.3486 0.8348 −0.0233 −0.1271 0.0073
ngtdm_Busyness 0.4294 0.0641 0.8060 −0.0878 −0.0794 −0.3625
glrlm_RunLengthNonUniformity 0.4816 0.3195 0.8002 −0.0529 −0.0978 0.0134
glszm_GrayLevelNonUniformity 0.5336 0.2082 0.7940 −0.1188 −0.0498 0.0161
glszm_SizeZoneNonUniformity 0.6006 0.2651 0.7215 −0.1111 0.1159 0.0154
ngtdm_Coarseness −0.2220 −0.4302 −0.8433 0.0989 0.0884 0.0284
gldm_DependenceEntropy 0.3518 −0.0681 −0.0639 0.8075 −0.2398 0.0633
glcm_Imc2 −0.3730 −0.4027 −0.1867 0.7782 −0.1732 0.0332
glcm_Correlation −0.3608 −0.4180 −0.1697 0.7754 −0.1810 0.0262
glcm_MCC −0.4282 −0.4166 −0.1623 0.7278 −0.0994 0.0534
glrlm_RunEntropy −0.6000 0.4408 −0.0414 0.6393 −0.0722 0.0348
glcm_Imc1 0.5096 0.3610 0.1685 −0.7387 0.0644 −0.0498
glszm_SizeZoneNonUniformityNormalized 0.0740 −0.0755 −0.1558 −0.2664 0.8512 0.0057
glszm_SmallAreaEmphasis 0.0612 −0.1059 −0.1709 −0.2477 0.8350 0.0006
glszm_ZoneEntropy 0.2966 0.3205 0.3745 0.4128 −0.6544 0.0229
glcm_InverseVariance 0.1879 −0.0715 −0.0066 −0.0949 −0.7197 −0.0769
ngtdm_Complexity 0.4364 −0.0339 0.0612 −0.0194 0.1146 0.8798
ngtdm_Strength −0.1725 −0.4529 −0.5426 0.1537 0.1237 0.6254
Open in a new tab

Figure 2.

Figure 2.

Open in a new tab

Radiomic Factor 1 loadings for a subset of the 75 radiomic features in the following categories: gray-level size zone matrix (GLSZM), gray-level co-occurrence matrix (GLCM), gray-level run length matrix (GLRLM), gray-level dependence matrix (GLDM), and neighboring gray-tone difference matrix (NGTDM). This factor can be best interpreted by the 37 radiomic features with highest magnitude loadings denoted by the red box.

Table 6 presents adjusted correlations for radiomic factors. Radiomic Factor 1 was significantly correlated with BMI, L3 muscle density, and DXA lean and fat masses (all p<0.001). Radiomic Factor 2 was significantly associated with age (p=0.037), L3 muscle area (p<0.001), L3 muscle density (p<0.001), grip strength (p=0.011), and DXA appendicular lean mass (p=0.031). Correlations of Radiomic Factor 3 with BMI, L3 muscle area and DXA lean mass variables were significant (all p≤0.027). Radiomic Factor 4 was only correlated with BMI (p=0.014) and DXA total body fat mass (p=0.036). Radiomic Factor 5 was significantly correlated with BMI (p=0.026), L3 muscle area (p=0.026), and L3 muscle density (p=0.012). Significant results for Radiomic Factor 6 were only seen for correlations with DXA total body fat mass (p=0.018).

Table 6.

Correlations between L3 muscle radiomic factors 1-6 with demographics, L3 muscle area and density, muscle strength metrics and DXA body composition measures. Significant associations, defined as p<0.05, are shown in bold.

Muscle Radiomic Factors Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6
Demographics r p-value r p-value r p-value r p-value r p-value r p-value
  Age1 0.053 0.527 −0.173 0.037 0.016 0.847 0.058 0.487 0.030 0.715 0.103 0.213
  BMI (kg/m2)2 0.642 < .001 0.023 0.783 0.183 0.027 −0.202 0.014 −0.183 0.026 −0.152 0.066
CT Metrics 2
  L3 Muscle Area (cm2) 0.036 0.662 0.323 < .001 0.441 < .001 −0.129 0.119 −0.184 0.026 −0.046 0.577
  L3 Muscle Density (HU) −0.297 < .001 0.918 < .001 −0.120 0.146 −0.057 0.492 0.207 0.012 −0.016 0.850
Muscle Strength 2
  Knee Extensor Strength (Nm) 0.125 0.155 0.159 0.072 0.136 0.124 −0.026 0.766 0.069 0.434 −0.172 0.051
  Grip Strength (kg) 0.040 0.632 0.209 0.011 0.153 0.065 −0.024 0.775 −0.066 0.424 −0.052 0.533
DXA Variables 2
  Total Body Fat Mass (kg) 0.776 < .001 −0.090 0.278 0.019 0.822 −0.173 0.036 −0.161 0.051 −0.194 0.018
  Lean Body Mass (kg) 0.365 < .001 0.120 0.149 0.419 < .001 −0.109 0.188 −0.139 0.093 −0.042 0.610
  Appendicular Lean Mass (kg) 0.284 < .001 0.178 0.031 0.353 < .001 −0.066 0.428 −0.110 0.183 −0.031 0.706
Open in a new tab

Spearman rank correlation coefficients and p-values

1

Adjusted for BMI and sex

2

Adjusted for age and sex

Discussion

The INVEST in Bone Health population can be described as a cohort of mostly female, moderately active, older adults living with overweight or obesity27. For this cohort, we observed significant associations between BMI and musculoskeletal measures, with higher BMI correlating to higher muscle and IMAT areas and BMI associated with over half of our radiomic factors. In contrast, age was largely unrelated to our variables of interest. Men and women differed in terms of muscle size, cortical thickness, and few cortical vBMD measures but we did not find significant differences for trabecular vBMD measures. Machine learning CT-derived measures of muscle size showed strong correlations with total body weight, and DXA-derived total body and appendicular lean masses. CT-derived measures of IMAT size and muscle density were strongly related to DXA-derived total body fat mass. For the mid-thigh, muscle size was significantly related to grip and knee extensor strength. Radiomic measures of muscle texture were associated more with BMI, CT-derived muscle size and density, and DXA-derived fat and lean masses than age and muscle strength metrics.

We did not observe age to be significantly correlated with most CT-derived muscle or bone measures, including radiomic factors, but there was a suggestion of a negative association between muscle size and density with age which is consistent with literature28, 29. The lack of significance may stem from our small sample size (n=150) or the relatively younger age of our cohort (66.4±4.6 years), possibly limiting observable differences in muscle size and density within this population.

BMI, as an estimate of weight-related health status, was highly correlated with soft tissue measures, including muscle and fat, while measures of bone lacked significance. However, our data showed that across all regions, trabecular vBMD was positively correlated with BMI while cortical vBMD was negatively correlated with BMI. This trend is consistent with prior studies30, which have shown BMI-defined obesity is associated with higher trabecular vBMD but lower cortical vBMD.

Consistent with previous work12, CT-derived muscle area and density measures were significantly higher in males than females across all regions. Additionally, cortical thicknesses were higher in males but vBMD for all compartments did not differ between the sexes. Sex-based vBMD differences have been less prominent after age 6031 which may explain our findings, but these results should be interpreted carefully as our sample was largely female (75%) and therefore, may have limited our ability to detect differences by sex.

In our study, we found consistent associations between CT-derived measures of muscle and vBMD with our traditional measures of interest, including muscle strength and DXA-derived aBMD, lean mass, and fat mass metrics. DXA total body and appendicular lean masses, used as muscle approximations, showed strong, direct correlations with muscle areas at the trunk and thigh. Fat mass, assessed by DXA, was strongly correlated with measures indicative of muscle quality including IMAT area (direct correlations) and muscle density (indirect correlations). Lower extremity muscle strength was directly associated with the corresponding muscle area and density measures. Additionally, DXA aBMD and CT vBMD measures were directly and significantly correlated across all sites. Although DXA is the primary tool for diagnosing osteoporosis and sarcopenia, it is prone to age- and obesity-related errors that affect reliability32–35. Muscle strength also falls short in its ability to directly assess muscle quality36. In contrast to DXA and muscle strength, CT offers more specific assessments of muscle, capturing clinically relevant variations in both quantity and quality.

In our analysis of the INVEST in Bone Health cohort, we observed that while CT-derived IMAT areas showed no association with strength measures, CT-derived muscle areas did, underscoring a clear relationship between muscle quantity and strength. CT muscle areas were also associated with total body mass, DXA-derived lean body mass, and appendicular lean mass, indicating that although DXA measures approximate muscle through lean mass, they correlate strongly with direct muscle area. Additionally, we found that CT muscle density and IMAT areas, indicators of muscle quality, were strongly associated with DXA total body fat mass. These findings support a positive relationship between traditional body composition measures, such as DXA and strength, with CT-based assessments. Thus, when direct CT measurements of muscle quantity and quality are unavailable, more widely used clinical metrics can still effectively estimate muscle outcomes.

In some cases, CT revealed associations with muscle strength or BMI that were not evident with DXA, which may suggest certain limitations of DXA. DXA is known to underdiagnose conditions like sarcopenia and osteoporosis—key musculoskeletal concerns for older adults with obesity. In contrast, CT offers more precise assessments of musculoskeletal health and may enhance the detection and monitoring of these conditions. Integrating automated CT measurements, including radiomics, into routine care could enable more detailed tracking of musculoskeletal changes and support timely interventions. Additionally, as machine-learning techniques advance, CT-derived metrics hold significant potential for opportunistic assessments during routine scans, proactively identifying conditions that DXA might miss.

The present study relies on BMI, calculated as weight divided by height squared, to define individuals as living with overweight or obesity. This measure, used in many of the presented analyses, is a known limitation as BMI is increasingly viewed as an outdated tool. While BMI offers a quick method to categorize weight-related health risks, it fails to consider differences in muscle mass, fat distribution, bone density, or overall body composition. Despite these limitations, no solid alternative to BMI currently exists. Future research may benefit from more holistic approaches, incorporating waist circumference, body fat percentage, and assessments of physical fitness and metabolic health.

The analysis of muscle radiomics in the present study is unique as few studies have explored how these radiomic features relate to age, BMI, and other musculoskeletal measures. To date, these novel features have primarily been used to predict clinical outcome such as mortality in select populations. As measures of muscle heterogeneity37, radiomics could provide insight on age- and weight-related differences in muscle health beyond current assessment techniques. However, these features, quantified via data-characterization algorithms38, introduce a multi-dimensional level of complexity to CT-based muscle analyses making their interpretation challenging. Here, we employed factor analysis to simplify the data, and a detailed interpretation of the resulting factors is provided in the Supplementary File. Our findings indicate that radiomic features are associated with age, BMI, muscle strength, and body composition measures (CT and DXA). However, despite these associations, the clinical relevance of these radiomic features remains uncertain due to their complexity, highlighting the need for further research to clarify their significance.

Conclusions

In summary, our study within the INVEST in Bone Health trial, focusing on older, moderately active adults with overweight or obesity, identified BMI as a key factor associated with CT-derived muscle measurements. We also found strong correlations between CT-derived muscle and bone metrics and traditional measures like DXA. Uniquely, this study explored muscle radiomics and linked these complex texture features to age, BMI, and musculoskeletal measures, offering new insights beyond traditional assessment methods. Our results support using machine learning to automatically derive CT musculoskeletal measurements, including radiomics, for future research on age- and obesity-related musculoskeletal health.

Supplementary Material

Supinfo2
Supinfo1

What is already known about this subject?

  • Aging and obesity together can lead to conditions like sarcopenic or osteosarcopenic obesity, which harm mobility, increase disability, mortality, and reduce quality of life.

  • Current methods for assessing musculoskeletal health, like dual energy x-ray absorptiometry (DXA) and muscle strength tests, have limitations, as they can’t directly measure muscle mass or quality.

  • Analyses of computed tomography (CT) scans offers deeper insights for muscle and bone health in older adults with obesity.

What are the new findings in your manuscript?

  • In a mostly female cohort of older adults (66.4±4.6 yo) living with overweight or obesity, higher body mass index (BMI) was linked to higher muscle area and fat infiltration, with no correlations to bone mineral density or cortical thickness.

  • Automatic and semi-automatic CT measures correlated strongly with corresponding DXA measures for muscle and bone tissue.

  • Novel muscle radiomic factors were significantly associated with age, BMI, muscle strength, and body composition from CT and DXA.

How might your results change the direction of research or the focus of clinical practice?

  • Our machine-learning CT analysis methods provide direct measures of muscle and fat infiltration, enabling opportunistic assessment of musculoskeletal disorders.

  • Muscle radiomics analysis and its links to age, BMI, and musculoskeletal metrics are novel, with potential as future clinical endpoints.

  • Higher BMI related to larger muscle areas but also increased IMAT in our study, highlighting a need to reduce adiposity while preserving musculoskeletal health in adults with overweight or obesity.

Acknowledgements

The authors gratefully acknowledge all of participants and thank them for their involvement in our study. We would also like to recognize members of the INVEST in Bone Health study team as well as student and postdoctoral trainees who have contributed to the INVEST in Bone Health project.

Computations for the present work were performed using the Wake Forest University (WFU) High Performance Computing Facility, a centrally managed computational resource available to WFU researchers including faculty, staff, students, and collaborators.

Funding:

This work was supported by the National Institute on Aging [Grants No. R01AG059186 (KMB); K25 AG058804 (AAW), F31 AG086010 (SDL), and P30 AG021332]. Jason Pharmaceuticals, a wholly owned subsidiary of Medifast, Inc. made an in-kind product donation for the meal replacements used in this study. Additionally, services and facilities of the Clinical Research Unit and Translational Imaging Program used for this work were funded by the National Center for Advancing Translational Sciences (NCATS), National Institutes of Health, through Grant Award Number UL1TR001420. The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Footnotes

Clinical Trial Registration: NCT04076618

Conflict of Interest Disclosure: The authors declared no conflicts of interest.

Data Sharing Plan

The current study focuses on baseline data from the INVEST in Bone Health clinical trial. As of April 2024, the study completed all data collection with analyses and dissemination of primary study results ongoing. Full trial results, including intervention-based outcomes, will be reported on ClinicalTrials.gov (NCT04076618) in the year following study completion. Data provided in this manuscript are available upon reasonable request.

References

  • 1.Oliveros E, Patel H, Kyung S, Fugar S, Goldberg A, Madan N, Williams KA. Hypertension in older adults: Assessment, management, and challenges. Clinical Cardiology. 2020;43(2):99–107. doi: 10.1002/clc.23303. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Alajlouni DA, Bliuc D, Tran TS, Blank RD, Center JR. Muscle strength and physical performance contribute to and improve fracture risk prediction in older people: A narrative review. Bone. 2023;172:116755. doi: 10.1016/j.bone.2023.116755. [DOI] [PubMed] [Google Scholar]
  • 3.Koliaki C, Dalamaga M, Liatis S. Update on the Obesity Epidemic: After the Sudden Rise, Is the Upward Trajectory Beginning to Flatten? Current Obesity Reports. 2023;12(4):514–27. doi: 10.1007/s13679-023-00527-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Ilich JZ, Kelly OJ, Inglis JE. Osteosarcopenic Obesity Syndrome: What Is It and How Can It Be Identified and Diagnosed? Current Gerontology and Geriatrics Research. 2016;2016:7325973. doi: 10.1155/2016/7325973. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Völgyi E, Tylavsky FA, Lyytikäinen A, Suominen H, Alén M, Cheng S. Assessing Body Composition With DXA and Bioimpedance: Effects of Obesity, Physical Activity, and Age. Obesity. 2008;16(3):700–5. doi: 10.1038/oby.2007.94. [DOI] [PubMed] [Google Scholar]
  • 6.Jensen B, Braun W, Geisler C, Both M, Klückmann K, Müller Manfred J, Bosy-Westphal A. Limitations of Fat-Free Mass for the Assessment of Muscle Mass in Obesity. Obesity Facts. 2019;12(3):307–15. doi: 10.1159/000499607. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Hangartner TN, Johnston CC. Influence of fat on bone measurements with dual-energy absorptiometry(0169-6009 (Print)). [DOI] [PubMed] [Google Scholar]
  • 8.Li GH-Y, Lee GK-Y, Au PC-M, Chan M, Li H-L, Cheung BM-Y, Wong IC-K, Lee VH-F, Mok J, Yip BH-K, Cheng KK-Y, Wu C-H, Cheung C-L. The effect of different measurement modalities in the association of lean mass with mortality: A systematic review and meta-analysis. Osteoporosis and Sarcopenia. 2021;7:S13–S8. doi: 10.1016/j.afos.2021.02.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Keaveny TM, Hoffmann Pf Fau - Singh M, Singh M Fau - Palermo L, Palermo L Fau - Bilezikian JP, Bilezikian Jp Fau - Greenspan SL, Greenspan Sl Fau - Black DM, Black DM. Femoral bone strength and its relation to cortical and trabecular changes after treatment with PTH, alendronate, and their combination as assessed by finite element analysis of quantitative CT scans(1523-4681 (Electronic)). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Löffler MA-O, Jacob A, Valentinitsch A, Rienmüller A, Zimmer C, Ryang YM, Baum T, Kirschke JS. Improved prediction of incident vertebral fractures using opportunistic QCT compared to DXA(1432-1084 (Electronic)). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Li X, Zhang Y, Xie Y, Lu R, Tao H, Chen S. Correlation Between Bone Mineral Density (BMD) and Paraspinal Muscle Fat Infiltration Based on QCT: A Cross-Sectional Study. Calcified Tissue International. 2022;110(6):666–73. doi: 10.1007/s00223-022-00944-6. [DOI] [PubMed] [Google Scholar]
  • 12.Graffy PM, Liu J, Pickhardt PJ, Burns JE, Yao J, Summers RM. Deep learning-based muscle segmentation and quantification at abdominal CT: application to a longitudinal adult screening cohort for sarcopenia assessment. The British Journal of Radiology. 2019;92(1100):20190327. doi: 10.1259/bjr.20190327. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Miller RM, Beavers DP, Cawthon PM, Crotts C, Fanning J, Gerosa J, Greene KA, Hsieh KL, Kiel J, Lawrence E, Lenchik L, Lynch SD, Nesbit BA, Nicklas BJ, Weaver AA, Beavers KM. Incorporating Nutrition, Vests, Education, and Strength Training (INVEST) in Bone Health: Trial Design and Methods. Contemp Clin Trials. 2021;104:106326. Epub 2021/02/26. doi: 10.1016/j.cct.2021.106326. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Stewart AL, Mills Km Fau - King AC, King Ac Fau - Haskell WL, Haskell Wl Fau - Gillis D, Gillis D Fau - Ritter PL, Ritter PL. CHAMPS physical activity questionnaire for older adults: outcomes for interventions(0195-9131 (Print)). [DOI] [PubMed] [Google Scholar]
  • 15.Beavers KM, Avery AE, Shankaran M, Evans WJ, Lynch SD, Dwyer C, Howard M, Beavers DP, Weaver AA, Lenchik L, Cawthon PM. Application of the D3-creatine muscle mass assessment tool to a geriatric weight loss trial: A pilot study. Journal of Cachexia, Sarcopenia and Muscle. 2023;n/a(n/a). doi: 10.1002/jcsm.13322. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Nicklas BJ, Chmelo E, Delbono O, Carr JJ, Lyles MF, Marsh AP. Effects of resistance training with and without caloric restriction on physical function and mobility in overweight and obese older adults: a randomized controlled trial(1938-3207 (Electronic)). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Crawley E, Evans W, Owen G. A theoretical analysis of the accuracy of single-energy CT bone mineral measurements. Physics in Medicine and Biology. 1988;33:1113–27. doi: 10.1088/0031-9155/33/10/002. [DOI] [PubMed] [Google Scholar]
  • 18.Bodden J, Dieckmeyer M, Sollmann N, Burian E, Rühling S, Löffler MT, Sekuboyina A, El Husseini M, Zimmer C, Kirschke JS, Baum T. Incidental vertebral fracture prediction using neuronal network-based automatic spine segmentation and volumetric bone mineral density extraction from routine clinical CT scans(1664-2392 (Print)). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Treece G, Gee A. Cortical Bone Mapping: Measurement and Statistical Analysis of Localised Skeletal Changes(1544-2241 (Electronic)). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Barnard R, Tan J, Roller B, Chiles C, Weaver AA, Boutin RD, Kritchevsky SB, Lenchik L. Machine Learning for Automatic Paraspinous Muscle Area and Attenuation Measures on Low-Dose Chest CT Scans. Acad Radiol. 2019;26(12):1686–94. Epub 2019/07/22. doi: 10.1016/j.acra.2019.06.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Lenchik L, Barnard R, Boutin RD, Kritchevsky SB, Chen H, Tan J, Cawthon PM, Weaver AA, Hsu FC. Automated Muscle Measurement on Chest CT Predicts All-Cause Mortality in Older Adults From the National Lung Screening Trial. J Gerontol A Biol Sci Med Sci. 2021;76(2):277–85. Epub 2020/06/07. doi: 10.1093/gerona/glaa141. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Chen L-C, Zhu Y, Papandreou G, Schroff F, Adam H, editors. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation. Computer Vision – ECCV 2018; 2018 2018//; Cham: Springer International Publishing. [Google Scholar]
  • 23.Systems I, University WF. WFU High Performance Computing Facility 2021. Available from: 10.57682/g13z-2362. [DOI] [Google Scholar]
  • 24.van Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, Beets-Tan RGH, Fillion-Robin JC, Pieper S, Aerts H. Computational Radiomics System to Decode the Radiographic Phenotype(1538-7445 (Electronic)). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Chen X-D, Chen W-J, Huang Z-X, Xu L-B, Zhang H-H, Shi M-M, Cai Y-Q, Zhang W-T, Li Z-S, Shen X Establish a New Diagnosis of Sarcopenia Based on Extracted Radiomic Features to Predict Prognosis of Patients With Gastric Cancer. Frontiers in Nutrition. 2022;9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Lenchik L, Webber K, Orwoll E, Cauley JA, Cawthon P, Ensrud K. CT-DERIVED PARAVERTEBRAL MUSCLE RADIOMIC FEATURES ARE ASSOCIATED WITH MORTALITY IN THE MROS STUDY(2399-5300 (Electronic)). [Google Scholar]
  • 27.Stewart AL, Verboncoeur CJ, McLellan BY, Gillis DE, Rush S, Mills KM, King AC, Ritter P, Brown BW Jr., Bortz WM II. Physical Activity Outcomes of CHAMPS II: A Physical Activity Promotion Program for Older Adults. The Journals of Gerontology: Series A. 2001;56(8):M465–M70. doi: 10.1093/gerona/56.8.M465. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Johannesdottir F, Allaire B, Anderson DE, Samelson EJ, Kiel DP, Bouxsein ML. Population-based study of age- and sex-related differences in muscle density and size in thoracic and lumbar spine: the Framingham study. Osteoporosis International. 2018;29(7):1569–80. doi: 10.1007/s00198-018-4490-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Anderson DE, D’Agostino JM, Bruno AG, Demissie S, Kiel DP, Bouxsein ML. Variations of CT-Based Trunk Muscle Attenuation by Age, Sex, and Specific Muscle. The Journals of Gerontology: Series A. 2013;68(3):317–23. doi: 10.1093/gerona/gls168. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Sukumar D, Schlussel Y, Riedt CS, Gordon C, Stahl T, Shapses SA. Obesity alters cortical and trabecular bone density and geometry in women. Osteoporosis International. 2011;22(2):635–45. doi: 10.1007/s00198-010-1305-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Zhang X, Hua T, Zhu J, Peng K, Yang J, Kang S, Xu T, Hu J, Tang G. Body compositions differently contribute to BMD in different age and gender: a pilot study by QCT. Archives of Osteoporosis. 2019;14(1):31. doi: 10.1007/s11657-019-0574-5. [DOI] [PubMed] [Google Scholar]
  • 32.Kelleher JL, Beavers DP, Henderson RM, Yow D, Crotts C, Kiel J, Nicklas BJ, Beavers KM. Weighted Vest Use during Dietary Weight Loss on Bone Health in Older Adults with Obesity. LID - 210 [pii] LID - 10.4172/2329-9509.1000210 [doi](2329–9509 (Electronic)). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Beavers KM, Walkup MP, Weaver AA, Lenchik L, Kritchevsky SB, Nicklas BJ, Ambrosius WT, Stitzel JD, Register TC, Shapses SA, Marsh AP, Rejeski WJ. Effect of Exercise Modality During Weight Loss on Bone Health in Older Adults With Obesity and Cardiovascular Disease or Metabolic Syndrome: A Randomized Controlled Trial(1523–4681 (Electronic)). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Beavers KM, Ambrosius WT, Rejeski WJ, Burdette JH, Walkup MP, Sheedy JL, Nesbit BA, Gaukstern JE, Nicklas BJ, Marsh AP. Effect of Exercise Type During Intentional Weight Loss on Body Composition in Older Adults with Obesity. Obesity. 2017;25(11):1823–9. doi: 10.1002/oby.21977. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Madrid DA, Beavers KM, Walkup MP, Ambrosius WT, Rejeski WJ, Marsh AP, Weaver AA. Effect of exercise modality and weight loss on changes in muscle and bone quality in older adults with obesity. Experimental Gerontology. 2023;174:112126. doi: 10.1016/j.exger.2023.112126. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Chen L, Nelson DR, Zhao Y, Cui Z, Johnston JA. Relationship between muscle mass and muscle strength, and the impact of comorbidities: a population-based, cross-sectional study of older adults in the United States. BMC Geriatrics. 2013;13(1):74. doi: 10.1186/1471-2318-13-74. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Lenchik L, Mazzoli V, Cawthon PM, Hepple RT, Boutin RD. Muscle Steatosis and Fibrosis in Older Adults, From the AJR Special Series on Imaging of Fibrosis. American Journal of Roentgenology. 2023:1–13. doi: 10.2214/AJR.23.29742. [DOI] [PubMed] [Google Scholar]
  • 38.Mayerhoefer ME, Materka A, Langs G, Häggström I, Szczypiński P, Gibbs P, Cook G. Introduction to Radiomics(1535-5667 (Electronic)). [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supinfo2
NIHMS2050364-supplement-Supinfo2.docx (14.1KB, docx)
Supinfo1
NIHMS2050364-supplement-Supinfo1.docx (184.4KB, docx)

Data Availability Statement

The current study focuses on baseline data from the INVEST in Bone Health clinical trial. As of April 2024, the study completed all data collection with analyses and dissemination of primary study results ongoing. Full trial results, including intervention-based outcomes, will be reported on ClinicalTrials.gov (NCT04076618) in the year following study completion. Data provided in this manuscript are available upon reasonable request.

RESOURCES