Population‐based cohort imaging: skeletal muscle mass by magnetic resonance imaging in correlation to bioelectrical‐impedance analysis

Abstract Background Skeletal muscle mass is subjected to constant changes and is considered a good predictor for outcome in various diseases. Bioelectrical‐impedance analysis (BIA) and magnetic resonance imaging (MRI) are approved methodologies for its assessment. However, muscle mass estimations by BIA may be influenced by excess intramuscular lipids and adipose tissue in obesity. The objective of this study was to evaluate the feasibility of quantitative assessment of skeletal muscle mass by MRI as compared with BIA. Methods Subjects from a population‐based cohort underwent BIA (50 kHz, 0.8 mA) and whole‐body MRI including chemical‐shift encoded MRI (six echo times). Abdominal muscle mass by MRI was quantified as total and fat‐free cross‐sectional area by a standardized manual segmentation‐algorithm and normalized to subjects' body height2 (abdominal muscle mass indices: AMMIMRI). Results Among 335 included subjects (56.3 ± 9.1 years, 56.1% male), 95 (28.4%) were obese (BMI ≥ 30 kg/m2). MRI‐based and BIA‐based measures of muscle mass were strongly correlated, particularly in non‐obese subjects [r < 0.74 in non‐obese (P < 0.001) vs. r < 0.56 in obese (P < 0.001)]. Median AMMITotal(MRI) was significantly higher in obese as compared with non‐obese subjects (3246.7 ± 606.1 mm2/m2 vs. 2839.0 ± 535.8 mm2/m2, P < 0.001, respectively), whereas the ratio AMMIFat‐free/AMMITotal (by MRI) was significantly higher in non‐obese individuals (59.3 ± 10.1% vs. 53.5 ± 10.6%, P < 0.001, respectively). No significant difference was found regarding AMMIFat‐free(MRI) (P = 0.424). In analyses adjusted for age and sex, impaired glucose tolerance and measures of obesity were significantly and positively associated with AMMITotal(MRI) and significantly and inversely with the ratio AMMIFat‐free(MRI)/AMMITotal(MRI) (P < 0.001). Conclusions MRI‐based assessment of muscle mass is feasible in population‐based imaging and strongly correlated with BIA. However, the observed weaker correlation in obese subjects may explain the known limitation of BIA in obesity and promote MRI‐based assessments. Thus, skeletal muscle mass parameters by MRI may serve as practical imaging biomarkers independent of subjects' body weight.


Introduction
Skeletal muscle, as one of the largest body compartments in adults, has a fundamental influence on health. It determines the physical condition, provides mobility and plays a key role in hormone and glucose metabolism and homeostasis. Therefore, skeletal muscle is a cardinal phenotypic marker and may serve as a good predictor for outcome in numerous health and disease states as well as an ideal target for health preservation and/or improvement. 1,2 Skeletal muscle is subject to significant changes in mass, composition and function with ageing and in various diseases. 3 These changes of morphology and physiology are based on inflammation, hormonal dysregulation, denervation of motor units, and intramuscular lipid deposition. 4 Major consequences are not only mobility limitations and physical disability with an increased risk of falls, injuries, and hospitalization but also cardiometabolic disorders with consecutively increased adverse health-events and all-cause mortality. 1,2,5 Due to ongoing demographic transition with progressive ageing and weight gain of the population, disease-related and age-related deterioration of skeletal muscle is an important public health issue. 4 Thus, simple, practicable and valid research, screening, and diagnosing tools as well as robust biomarkers for the assessment of relevant muscle properties are of high clinical importance.
Currently, bioelectrical-impedance analysis (BIA) is a commonly used methodology for body composition estimations and the rough assessment of skeletal muscle mass. Its strength is based on the easy and safe way to use and also on its cost-efficiency compared with other methods. However, BIA features a large individual prediction error for the estimation of skeletal muscle mass, because its reliability is influenced by multiple factors [e.g. device-related factors, such as intra-instrumental/inter-instrumental variability), operator-related factors (such as intra-operator/inter-operator variability), subject-related factors (such as age, ethnicity, body weight and temperature, fasting and hydration state, and recent activity), and environment-related factors (such as room temperature)]. Also, BIA is not able to accurately determine intramuscular fatty and fibrous components. 6 In contrast, magnetic resonance imaging (MRI) with its lack of ionizing radiation, excellent soft tissue contrast, and high spatial resolution provides the possibility of simultaneous assessment of mass and other morphological and textural features and is therefore particularly suited with regard to large co-hort settings. Furthermore, it is also sensitive to very small changes in muscle mass and allows concomitant evaluation of surrounding structures relevant for muscle activity and performance, such as bones, joints, or tendons. 6,7 Therefore, our hypothesis is that MRI is a feasible alternative to other accepted modalities of skeletal muscle assessment and that muscle mass parameters as assessed by MRI may serve as feasible imaging biomarkers in clinical and research settings independent of subjects' body weight. Aim of the present study was to evaluate the feasibility of quantitative assessment of total and fat-free muscle mass by MRI using a standardized manual segmentation approach in non-obese and obese subjects from the general population. This segmentation approach allows for the simultaneous assessment of additional muscle parameters (such as total, intramyocellular, and extramyocellular fat content), complementing quantitative and qualitative analyses.

Study design and population
Subjects were derived from the KORA-FF4 study (2013-2014, n = 2279), the second follow-up of a population-based survey within the Cooperative Health Research in the Region of Augsburg (KORA) survey. The design of the KORA studies has been described in detail previously. 8 In brief, a comprehensive health assessment was prospectively performed for all subjects. Additionally, 400 subjects underwent whole-body MRI according to previously described inclusion and exclusion criteria. 8 Median time to MRI examination was 33 days (IQR: 24-45 days) after the health assessment. The study was approved by the ethics committee of the Bavarian Chamber of Physicians, Munich, Germany, and the local institutional review board of the Ludwig-Maximilians University Munich, Germany. The study complies with the Declaration of Helsinki, including written informed consent of all participants.

Anthropometry
The body mass index (BMI) was calculated as weight in kg divided by body height squared in square meters, with body weight and height both measured at the study centre. Waist circumference was measured at the smallest abdominal circumference or, in obese subjects, in the midpoint of the lowest rib and the upper margin of the iliac crest. Hip circumference was determined at the most protruding part of the hips to the nearest 1 mm. Obesity was defined according to the WHO-definition with a BMI of 30 kg/m 2 as cut-off value. 9

Bioelectrical-impedance analysis
Whole-body BIA-scans were acquired using a body impedance analyser (BIA 2000-S, Data-Input, Pöcking, Germany) with an operating frequency of 50 kHz at 0.8 mA to determine total body fat mass, lean body mass, and appendicular muscle mass in kilograms as well as the correspondent indices in kg normalized to subjects' body height squared. Therefore, ohmic resistance was measured between the dominant hand wrist and dorsum and the dominant foot angle and dorsum in supine position.
Muscle mass in kg was then calculated by the following equation according to Janssen et al. 10  with body height in centimeter, resistance in Ω, for gender: male = 1 and female = 0 and age in years. 3,10 Imaging protocol and data acquisition Magnetic resonance imaging examinations were performed in supine position on a 3-Tesla Magnetom Skyra (Siemens Healthineers, Erlangen, Germany) using an 18-channel body surface coil in combination with a table-mounted spine matrix coil. The complete imaging protocol and technical specificities have been described in detail elsewhere. 8 For muscle segmentation, a T2*-corrected, multi-echo 3Dgradient-echo Dixon-based sequence (multi-echo Dixon) of the upper abdomen providing a coverage of approximately 40 to 50 cm with the following parameters was used (time to repetition (TR): 8.90 ms; time to echo (TEs): 1.23, 2.46, 3.69, 4.92, 6.15, and 7.38 ms; flip angle 4°, readout echo bandwidth 1080 Hz/pixel, matrix 256 × 256, slice thickness 4 mm). Data were acquired during a single breath-hold of 15 s. The post-processing algorithm using the Software MR LiverLab (Version VD13, Siemens Healthineers, Cary, USA) automatically calculated proton-density fat-fraction maps as DICOM-files, which were used to simultaneously assess muscle mass and fat content.

Image analysis and muscle segmentation
The DICOM-files were implemented into the Software MITK (Version 2015.5.2, German Cancer Research Center, Heidelberg, Germany). Two independent observers blinded to any covariates performed muscle segmentation. Details of the applied segmentation approach have been described previously. 11 In brief, each muscle compartment (both the right and left psoas major, quadratus lumborum, rectus abdominis, and autochthonous back muscles) was manually segmented according to standardized anatomical landmarks on one axial slice at the level of the lower endplate of the L3 vertebral body, because recent studies demonstrated that level L3 is a good surrogate for the entire lumbar spine and that skeletal muscle cross-sectional area at this level is a reliable method for the determination of sarcopenia. 12,13 If L3 vertebra was not imaged, the most caudal possible axial slice was selected. Subjects with significant image artefacts on all levels were excluded from the analysis. The complete manual segmentation and post-processing procedure took an average 10 min.
Interobserver and intraobserver reproducibility of the segmentation algorithm was assessed in a subset of 50 randomly selected subjects, being excellent for all included muscle compartments with only minor absolute and relative differences [intraclass correlation (ICC): 0.93-0.97, 31.0 ± 44.7 mm 2 , 2.7 ± 3.9%; ICC: 0.96-0.98, 5.5 ± 25.3 mm 2 , 0.5 ± 2.3%; respectively]. 11 Total and fat-free muscle mass Muscle compartments were segmented as described earlier using dedicated, anatomical landmarks defining the exact muscle boundaries. Muscle cross-sectional area (CSA) in mm 2 was then used as surrogate for MRI-based muscle mass. Total mass was defined as the complete muscle CSA Total comprised by its muscle fascia, including macroscopically visible, intermyocellular-intrafascial fatty septa. For the determination of fat-free muscle mass (CSA Fat-free ), the segmented compartments were post-processed using a semi-automated, inhouse application (Matlab_R2017a, The MathWorks, Inc., Massachusetts, USA) ( Figure 1). Given the presumptions that even myocytes with a high amount of intramyocellular lipids do not feature an intensity value greater than approximately 200 (corresponding to 20% fat content) in proton-density fatfraction maps and that every voxel with an intensity value greater than 200 therefore contains extramyocellular adipose tissue, a former approved threshold value of 200 was set to quantify those voxels that solely comprise myocytes with intramyocellular lipids, excluding extramyocellular, adipose tissue. 14 Fat-free mass was accordingly calculated by the following equation: Total and fat-free muscle mass were then normalized to subjects' body height squared calculating the correspondent indices in mm 2 /m 2 indicated as total and fat-free abdominal muscle mass index (AMMI Total and AMMI Fat-free , respectively). 15 Furthermore, the ratio AMMI Fat-free /AMMI-Total in % was calculated to indicate the amount of functionally contractable muscle tissue within the whole muscle compartments.

Myosteatosis
Total myosteatosis was quantified as mean proton-density fat-fraction in percentage (PDFF muscle ) in the same segmented muscle compartments at level L3 vertebra. 16

Visceral and subcutaneous adipose tissue
Visceral and subcutaneous adipose tissue (VAT and SAT, respectively) as abdominal adipose tissue compartments were segmented and quantified in square centimeters by a Skeletal muscle mass by MRI semi-automated algorithm based on fuzzy-clustering on one axial slice at the level of the umbilicus. 17,18 Other covariates To determine the glycaemic status, a 75 g oral glucose tolerance test was performed for all subjects not yet diagnosed with diabetes mellitus. According to the WHO definition, subjects were classified with an impaired glucose metabolism either with established type 2 diabetes mellitus (T2DM) or prediabetes and as healthy controls. Hypertension was determined as systolic blood pressure ≥140 mmHg and/or diastolic blood pressure ≥90 mmHg or current intake of antihypertensive medication, given that the subjects were aware of being hypertensive. Alcohol consumption and smoking status were classified by self-report as no (0 g/day) or any alcohol consumption (≥0.1 g/day) and never-smoker and ex-smoker or current (regular/sporadic) smoker. Nutritional supply was evaluated with a 24 h food list and food frequency questionnaire. Regarding physical activity, subjects were categorized as physically active (regular physical activity ≥1 h/week) or physically inactive (irregular physical activity <1 h/week, almost no or no physical activity). Routinely intake of medication was generally categorized according to most recent guidelines. Standardized laboratory tests were used to determine important blood values. Dyslipidaemic changes of blood lipids were defined as increased levels of triglycerides, total cholesterol and LDL and decreased levels of HDL. The presence of somatic (musculoskeletal) symptoms, such as back pain and pain in arms, legs or joints was recorded in addition.

Statistical analysis
Demographics and characteristics of the study population are presented as arithmetic mean ± standard deviation or median with 1st and 3rd quartile for continuous variables and absolute counts with percentages for categorical variables. Differences in mean values or counts between non-obese and obese subjects were assessed by t test (quantitative data) or χ 2 -test (qualitative data). Correlations of MRI-based and BIA-based measurements of muscle mass as well as with demographics were evaluated by scatter plots and Pearson's correlation coefficients. Explorative associations of MRI-based measurements of muscle mass with demographics and measures of obesity and adipose tissue were determined by linear regression adjusted for age and gender. Outcomes and covariates were standardized by subtracting the mean and dividing by standard deviation. Statistical significance was indicated by P values <0.05. Statistical analysis was performed using R V3.4.1 (R Core Team, www.r-project. org, 2017).

Study population
Among 400 subjects who underwent whole-body MRI, 335 subjects (83.7%) were included in this analysis. Fifty-seven subjects (14.3%) were excluded due to insufficient image quality (e.g. fat-water swapping artefacts due to field inhomogeneities, and motion/breathing artefacts) or because they did not complete the imaging protocol and eight subjects (2%) were subsequently excluded due to missing values in any of the covariates. Demographics and characteristics of the study population are provided in Table 1.
In general, included subjects were predominantly middle-aged men (mean age: 56.3 ± 9.1 years; male gender: 56.1%) with 95 subjects (28.4%) being classified as obese based on a BMI ≥ 30 kg/m 2 . Overall, obese subjects featured a more distinct cardiometabolic risk profile having a higher prevalence of impaired glucose metabolism, hypertension, and dyslipidaemia and being significantly less physically active. Furthermore, obese subjects had significantly lower levels of vitamin D, significantly higher consumption rates of alcohol and nicotine and a significantly higher prevalence of musculoskeletal pain symptoms (all P < 0.04; Table 1).

Body composition and muscle mass
Detailed measurements of body composition by anthropometry, BIA, and MRI are shown in Table 2. Non-obese subjects had a significantly lower waist and hip circumference in anthropometry and based on BIA significantly lower indices of total body fat mass, lean body, and appendicular muscle mass as well as calculated SMI BIA (all P < 0.001).
Based on MRI, mean total and fat-free abdominal skeletal muscle mass of the entire sample by MRI was 8759.2 ± 2143 mm 2 and 5099.8 ± 1749.2 mm 2 , respectively. AMMI Total was lowest in female subjects with normal body weight and highest in obese male subjects (non-obese female subjects: 2495.3 ± 389.6 mm 2 /m 2 , obese male subjects: 3501.1 ± 532.9 mm 2 /m 2 , P < 0.001). No significant difference was found regarding AMMI Fat-free between non-obese and obese subjects (1695.9 ± 478.3 mm 2 /m 2 , 1743.7 ± 527.8 mm 2 /m 2 , respectively, P = 0.424). Concerning the ratio of normalized fat-free to total abdominal skeletal muscle mass AMMI Fat-free /AMMI Total and in contrast to AMMI Total , normal weight male and female subjects had a significantly higher portion of fat-free muscle tissue compared with obese, male and female subjects (male: 61.4 ± 9.7% and 56.3 ± 11.4%, P = 0.007, female: 56.6 ± 10.1% and 50.0 ± 8.4%, P < 0.001) (Figure 2). Furthermore, regarding the main body fat compartments by MRI, obese subjects with a BMI ≥ 30 kg/m 2 showed significantly higher amounts of VAT, SAT, and PDFF muscle compared with normal weights (P < 0.001).

Predictors of muscle mass
Associations of muscle mass by MRI with demographics and cardiometabolic risk factors are provided in Table S1 and Figure S2. Regarding muscle composition, higher levels of PDFF muscle indicating myosteatosis are associated with a significantly lower AMMI Fat-free and ratio AMMI Fat-free /AMMI-Total , respectively (P < 0.001).
In explorative analyses adjusted for age and gender, impaired glucose metabolism, measures of dyslipidaemia, obesity, and adipose tissue (elevated triglyceride levels, BMI ≥ 30 kg/m 2 , higher waist and hip circumference, VAT and SAT) as well as hypertension were significantly and positively associated with AMMI Total (P < 0.01). Furthermore, a significantly positive association was found for AMMI Total and musculoskeletal pain symptoms (P = 0.002) as well as for AMMI Fat-free and total cholesterol, LDL and creatinine (P < 0.03). However, no significant association of muscle mass parameters by MRI with physical activity, nutrient supply, alcohol and nicotine consumption was found (P > 0.05).

Discussion
Skeletal muscle has a fundamental influence on physical and metabolic health. Its characteristics may therefore serve as imaging biomarkers for the general and cardiometabolic health status for both clinical and research purposes. In this study, we evaluated a straight-forward approach to total and fat-free muscle mass by MRI complementing other, qual-itative imaging parameters of muscle composition, and compared it to one of the standard methods used for muscle quantification, that is, BIA. On the one hand, our results confirm the significant prediction error of muscle mass by BIA in obesity, with considerably stronger correlations of MRI-based and BIA-based measures in normal weight subjects. On the other hand, our results indicate different patterns of MRI-derived mass parameters between obese and non-obese subjects as well as specific associations with risk factors for cardiometabolic disease. Thus, MRI allows for a comprehensive assessment of different morphologic characteristics of muscle mass biomarkers, for example, in cardiometabolic risk stratification, independent of subjects' body weight.
Currently, there are several modalities available for body composition analysis and estimation/quantification of muscle mass, relying on different technologies and assessing different aspects of skeletal musculature. However, the most commonly applied methods have substantial limitations and disadvantages. First, anthropometry, as an indirect modality for muscle mass quantification, is generally limited by its susceptibility to significant, individual and obesity-related prediction errors with the tendency to overestimate muscle mass. 19,20 Accordingly, anthropometric and MRI-based measures showed no to only weak correlations in this study. Second, dual-energy X-ray absorptiometry (DEXA), which some authors consider as the reference standard for quantification of muscle mass, 6 is less sensitive to small changes compared with, for example, CT and MRI 21,22 and additionally based on ionizing radiation, thus limiting its feasibility. Third, cross-sectional imaging, such as CT, is indeed characterized by a considerably smaller error in quantifying total and fat-free skeletal muscle muss. However, the amount of ionizing radiation involved in whole-body CT is limiting its utility in clinical/preventive and also research settings. Also, MRI as another cross-sectional modality has several contraindications (e.g. metallic implants and claustrophobia), is more time consuming and costly compared with, for example, BIA and DEXA and requires a higher amount of technical expertise. Fourth, BIA lacks standardization of the technical procedure and its validity may be strongly influenced by different disease conditions, hydration status and physical activity in temporal relation with the measurement procedure. Thereby, BIA generally tends to overestimate lean body mass and consecutively underestimates total body fat mass in conditions with water retention, such as liver, renal, or heart failure, after exercise and most importantly in obese subjects. 6,23 In summary, there is no available technique that serves all the requirements for the precise and practical measurement of skeletal muscle mass until date. Each method has its own advantages and limitations, none is fully standardized, and therefore, none serves as gold standard at present.
In this study, measures of muscle mass assessed by MRI and BIA correlated considerably stronger in normal weight compared with obese subjects. Also, obese subjects featured comparatively higher values in BIA-derived measurements of lean body and appendicular muscle mass as well as calculated skeletal muscle mass index. Independent of subjects' actual muscle mass, BIA generally overestimates lean body and therewith muscle mass in obesity. Furthermore, BIA is unable to determine intramuscular lipids and adipose tissue, which might account for a large fraction of non-myocellular content in muscle tissue. According to this, significant differences were not found regarding AMMI Fat-free between obese and non-obese subjects by MRI, whereas in BIA, obese subjects featured significantly higher muscle mass parameters. Overall, these results confirm the well-known, significant prediction error of muscle mass parameters by BIA in obese subjects, which accounts primarily for the markedly lower correlation coefficients in subjects with a BMI ≥ 30 kg/m 2 . Thus, MRI might outperform BIA regarding quantification of skeletal muscle mass parameters specifically in obese subjects with a BMI ≥ 30 kg/m 2 , for example, in the assessment of sarcopenic obesity, an obese subtype with relatively low muscle mass. Another advantage of MRI is the possibility to additionally perform a morphological and qualitative evaluation of muscle tissue complementing mass measurements, such as the assessment of intramyocellular and extramyocellular fat content and their distribution patterns as well as further, textural analyses. Because mass alone does not predict muscular performance sufficiently, MRI may be particularly suited for muscle composition analyses determining further aspects of related to its functional properties. Given the high likelihood of a relationship between MRI-detectable muscle composition patterns, muscular performance and cardiometabolic disease, associations between these quantitative imaging findings with both clinical performance and outcome (e.g. in ageing and development of cardiometabolic disease) need to be further determined in large cohort studies. 24 Some limitations of this study should be taken into account. First, we did not compare our results to DEXA, which is considered as the current gold standard by some authors. However, previous studies have demonstrated the validity and reproducibility of the standardized, muscle quantification by multi-echo Dixon that we used in this study. 11 Second, muscle performance does not depend solely on mass. Besides distinct associations of cardiometabolic risk factors and MRI-derived measurements of muscle composition in this study, these imaging biomarkers should further be evaluated regarding their diagnostic and prognostic value and therefore rather be seen as a complement for conventional strength and endurance tests. Third, this study focused on the evaluation of muscle mass parameters by MRI and their correlation with BIA-based measurements in both non-obese and obese subjects. We did not interpret our results in reference to a standard population and did not aim to categorize subjects as sarcopenic or find a cut-off value for sarcopenia in this population. However, as it is generally accepted that an absolute muscularity below the 5th percentile indicates sarcopenia, 25 the applied approach should be further evaluated as a simple estimate of the quantitative index in this condition.
The assessment of muscle mass parameters by MRI based on a standardized segmentation algorithm is a feasible alternative to BIA in population-based imaging. The weaker correlation of BIA and MRI in obese subjects may explain the known limitation of BIA in obesity and further promote MRI-based assessments in large cohort settings in both non-obese and specifically in obese subjects. AMMI Fat-free and AMMI Total as well as their ratio as surrogates for muscle mass by MRI might serve as promising imaging biomarkers for the comprehensive assessment of skeletal muscle and whole-body composition analysis as well as for a more accurate, cardiometabolic risk stratification, independent of subjects' body weight. AMMI, abdominal muscle mass index; BIA, bioelectrical-impedance analysis; BMI, body mass index; CSA, cross-sectional area; MRI, magnetic resonance imaging; PDFF, proton-density fat-fraction; SAT, subcutaneous adipose tissue; VAT, visceral adipose tissue.
Estimates are based on linear regression with standardized outcome and covariates. Standardization consisted of subtracting the mean and dividing by the standard deviation. † Skeletal muscle mass index derived by the following equation and normalized to subjects body height squared: Skeletal muscle mass (kg) = (body height 2 / resistance × 0.401) + (gender × 3.825) + (age × À0.071) + 5.102 (body height in cm, resistance in Ω, for gender: male = 1 and female = 0, age is in years).
Skeletal muscle mass by MRI Figure S1. Correlations of MRI-and BIA-based measurements of skeletal muscle mass in non-obese (light circle) and obese (dark triangle) subjects. Figure S2. Correlations of AMMITotal (A) and AMMIFat-free (B) with age, BMI, VAT and physical activity. Table S1. Associations between demographics, cardiometabolic risk factors and AMMI Total and AMMI Fat-free Table S2. Obesity-stratified associations of MRI-and BIA-based measurements of skeletal muscle muss.