Improved Strength Prediction Combining MRI Biomarkers of Muscle Quantity and Quality

Abstract

Muscle strength declines with aging at a faster rate compared with muscle mass, suggesting that not only muscle quantity but also muscle quality and architecture are age-dependent. This study tested the hypothesis that quantitative MRI (qMRI)-derived biomarkers of muscle quality (fractional anisotropy [FA], radial diffusivity [RD], axial diffusivity [AD], fat fraction [FF], and T₂ relaxation time) and architecture (fascicle length) could improve the prediction of skeletal muscle strength over muscle mass alone. We recruited 24 adults (12 female, age range 30–79 years). Muscle mass was estimated as the volume and cross-sectional area (CSA) of the quadriceps. FA, RD, and AD parameters, together with fascicle length for the rectus femoris (RF) and vastus lateralis (VL), were derived from diffusion tensor imaging (DTI), and muscle-T₂ was calculated from a multi-echo spin echo sequence. FF was determined using the Dixon approach. CSA values were combined with FF to calculate the lean CSA. Isometric, eccentric, and concentric knee extension torques were measured for the left and right leg using an isokinetic dynamometer. The univariable assessment of torque was performed using a linear regression. The statistical significance of adding qMRI parameters to the torque prediction models was tested using a mixed-effect regression. The best univariable predictor of isometric, eccentric, and concentric torque was lean CSA. Adding FA, RF fascicle length, and VL fascicle length to the model improved the prediction of concentric torque compared with CSA alone. The addition of FA, T₂, RD, RF fascicle length, and VL fascicle length improved the prediction of eccentric torque over CSA alone. The addition of FF was not significant within the model. Our results confirmed the hypothesis that the inclusion of qMRI parameters of muscle composition and architecture leads to higher R² coefficients for the prediction of muscle strength compared with models solely based on muscle quantity. These observations support the utility of qMRI for future research on sarcopenia prediction and management.

1 |. Introduction

Muscle strength is critical for mobility and human health. Sarcopenia, an age-related condition characterized by a progressive and accelerated loss of muscle mass and strength [1], represents a major cause of morbidity and mortality in the aging population [2, 3] and has a tremendous impact on healthcare costs due to the increased odds of being hospitalized [4], longer and more costly hospitalizations [5], and the need for longer-term care. Sarcopenia significantly contributes to the decline in physical function observed in older adults, as it impairs activities of daily living such as walking, rising from a chair, or climbing stairs [6, 7]. This decline in functionality also makes older adults more susceptible to falls, fractures, and prolonged immobility [8]. Additionally, sarcopenia often coexists with other conditions such as osteoporosis, malnutrition, systemic inflammation, osteoarthritis [9], and depression [10], which further increase health risks and accelerate physical decline. As populations age globally, the burden of sarcopenia on healthcare systems is expected to rise significantly [11], underscoring the need for early identification and targeted interventions.

Despite its high incidence (up to 51% for men and 31% for women in nursing homes [12]), the identification of early signs of sarcopenia, prior to irreversible loss in muscle function and strength, remains challenging, leading to a suboptimal management of sarcopenic patients [13]. This is due to the fact that while sarcopenia is defined as a pathological condition, it is inextricably linked to the aging process. Therefore, it is of paramount importance to develop biomarkers that can differentiate between sarcopenia and “healthy aging.”

Muscle mass is the strongest predictor of loss of strength in older adults [14], but it cannot fully explain the functional decline in muscle strength across the lifespan. In fact, skeletal muscle strength is known to decrease with age at a rate that exceeds the concomitant loss in muscle mass [15]. This clearly indicates that other aspects, besides muscle quantity, contribute to the progressive weakness in older adults.

The mechanisms affecting muscle quality are both neurological and intrinsic skeletal muscle properties [16]. At the skeletal muscle level, muscle aging and sarcopenia are characterized by many compositional and structural changes that contribute to reduced strength and functional capacity. One of the hallmark features is the selective atrophy of type 2 (fast-twitch) muscle fibers [17]. In addition to fiber atrophy, sarcopenia is associated with infiltration of fat within the muscle—known as myosteatosis—which impairs muscle quality and contractile efficiency [18]. Chronic low-grade inflammation and mitochondrial dysfunction are also commonly observed and may contribute to the degradation of muscle proteins and reduced regenerative capacity [19]. Additionally, muscles in older adults undergo geometrical remodeling [20], such as changes in fascicle length. These micro and macroscopic aspects of muscle composition and architecture are collectively referred to as “muscle quality,” according to the revised consensus of the European Working Group on Sarcopenia in Older People (EWGSOP2) [21] Practically, muscle quality is often defined as the ratio between voluntary muscle strength and muscle size [22], but no consensus on the exact practical definition or assessment methods exists. Despite this lack of consensus, the importance of muscle quality is increasingly recognized, and it has been included in the diagnostic criteria of sarcopenia, together with muscle quantity and muscle strength [21]. Changes in muscle quality likely precede global and irreversible loss of muscle mass [23], contribute to decreased physical function, and together with the decrease in muscle quantity, collectively contribute to lower muscle strength and increased risk of falls [24]. The detection of these early changes in muscle quality could help identify older adults at risk of developing sarcopenia and allow the application of early and more effective interventions. Furthermore, the ability to detect several aspects of muscle quality simultaneously would allow for a more objective identification of the factors that contribute to muscle weakness and could provide specific targets for treatment.

Due to the complex and multifactorial nature of sarcopenia, it is unlikely that a single biomarker, such as muscle mass, could provide a complete assessment of the musculature. There is therefore growing interest in combining measures of muscle quantity and quality that better relate to muscle function and muscle strength.

Computed tomography (CT), dual-energy X-ray absorptiometry (DEXA), and ultrasound (US) can provide accurate monitoring of global tissue loss by measuring muscle cross-sectional area (CSA) [25, 26] and assessing fat infiltration [27]. However, despite their robust performance in detecting global changes in muscle quantity [28], CT, DEXA, and US fail at identifying fundamental early changes in muscle quality [29, 30], such as cellular atrophy, inflammation, and 3D architectural remodeling.

Magnetic resonance imaging (MRI) is currently being established as the most powerful imaging tool in sarcopenia research, as it is the only imaging modality that can simultaneously monitor muscle quantity [25], muscle composition [25, 26], and fiber architecture [31]. Conventional MRI primarily produces high-resolution images that allow for visual assessment of muscle anatomy. Quantitative MRI (qMRI) enhances conventional MRI by providing objective measurements of muscle composition and architecture [32]. One of the most often utilized qMRI techniques is chemical shift MRI, combined with Dixon processing, which allows decoupling of the MRI signal originating from water and fat [33], and can therefore be used to quantify fatty infiltration [27]. Fatty infiltration derived from Dixon has been shown to be a strong predictor of muscle strength [34], but its association with function and strength in the context of muscle aging is less clear.

Another useful biomarker of muscle health is water T₂ of muscle mass (to distinguish it from the global T₂ of muscle and fat combined), which can provide an assessment of water content and tissue inflammation.

Diffusion tensor imaging (DTI) is an MRI-based technique that exploits the random motion of water molecules inside and around muscle cells to study muscle microstructure, and is therefore a promising tool to study early signs of cellular atrophy. In particular, radial diffusivity (RD) and fractional anisotropy (FA), two scalar measures obtained from DTI, have been linked to muscle strength [35], suggesting sensitivity to muscle fiber size and muscle fiber type [36]. However, these studies were performed on young subjects, and it is unclear whether the findings can be generalized to people of different ages. Besides microstructure, DTI can also be used to study other muscle characteristics affected by aging, such as fascicle length. In fact, DTI can be leveraged to reconstruct the macroscopic arrangement of fascicles within a muscle, and to derive estimates of fascicle length [37], which can be used to infer functional properties of muscle tissue [31].

Overall, due to its ability to assess several aspects of muscle quantity and quality objectively, qMRI holds great potential as a tool to investigate the causes of muscle weakness in older adults and to detect early signs of sarcopenia. This suggests that combining MRI measures of skeletal muscle quantity and quality can improve the prediction of skeletal muscle function. However, as multiple aspects of muscle quantity and quality are affected by aging, it is important to assess their relative contribution to age-related muscle weakness. In particular, considering the known fundamental role of muscle mass in overall muscle function and strength, useful qMRI biomarkers of muscle quality should be able to predict muscle strength independently of muscle mass.

In this work, we studied the quadriceps muscles of healthy subjects across a large age range using qMRI techniques, including Dixon, DTI, and T₂ mapping, and compared them with muscle strength as measured by knee extension torque. Given the proven strength of MRI in assessing several aspects of muscle quality, we hypothesize that the combination of MRI-derived muscle CSA and qMRI biomarkers can improve the prediction of knee extension torque over muscle CSA alone.

2 |. Materials and Methods

2.1 |. Subjects

This study was approved by the Institutional Review Board at Stanford University, and all subjects signed informed consent prior to participation. Adult participants aged between 30 and 80 years were recruited. Subjects were excluded if they had contraindications to MRI scans or were pregnant. Additional exclusion criteria included: hormonal supplementation, osteoporosis (self-reported), osteoarthritis in the lower extremities (self-reported), body mass index (BMI) > 35 kg/m², use of corticosteroids in the past 2 years, diabetes, history of trauma in the lower extremities, and underlying muscle disease, including myopathies and sarcopenia. Subjects were asked not to perform strenuous physical activities the day prior to the visit. All measurements were performed on the same day, with MRI immediately following the measurements of strength, except for one subject, for whom the MRI was performed 2 months after the functional assessment. Physical activity level was assessed using the International Physical Activity Questionnaire (IPAQ) long form [38].

2.2 |. MRI Acquisition

All subjects were scanned at 3T (Signa Premier, GE Healthcare) using a 16-element anterior coil (AirCoil, GE Healthcare) placed on the thigh and the 12-element built-in posterior array for signal reception. The participants were placed feet-first supine in the scanner with their lower legs supported by sandbags to minimize gross motion. For every subject, three different datasets were collected, in the following order: Dixon, DTI, and multi-echo spin echo (MESE), each one in two partially overlapping stacks. The amount of overlap between the two stacks was manually adjusted for each subject, and care was taken to achieve complete coverage of the upper leg. The parameters for the MRI acquisitions were the following: Dixon: multi-echo gradient-echo sequence, TR = 13.3 ms, TE₁ = 1.14 ms, echo train length (ETL) = 12, voxel size = 1.76 × 1.76 × 6 mm³, matrix size = 256 × 256 × 42, number of signal averages (NSA) = 1, flip angle (FA) = 5°, scan time = 2 min 57 s × 2 stacks. DTI: Stejskal–Tanner sequence combined with a single-shot EPI readout, TR = 3500 ms, TE = 43 ms, voxel size = 3.6 × 3.6 × 6 mm³, matrix size = 126 × 126 × 42, NSA = 3, scan time = 3 min 13 s × 2 stacks. We used water excitation pulses combined with the slice selection gradient reversal (SSGR) approach. Diffusion was encoded along 15 non-collinear directions with b-value b = 400 s/mm², Δ ~20 ms, together with three non-diffusion weighted volumes. MESE: TR = 4000 ms, 5.6 < TE₁ < 7.4 ms, ETL = 16, voxel size = 1.76 × 1.76 × 6 mm³, matrix size = 256 × 256, 21 slices with 6 mm gap, NSA = 1, scan time = 3 min 52 s × 2 stacks.

2.3 |. MRI Data Analysis

The multi-echo gradient-echo data were reconstructed using the IDEAL [39] algorithm to obtain water- and fat-only images. The left and right quadriceps muscle groups were manually segmented by the same trained individual (1 year of training) from these Dixon water images. Global muscle volume was calculated from these segmented masks, and global CSA was defined as the sum of the maximum CSA for every individual muscle (vastus lateralis [VL], vastus medialis, vastus intermedius, and rectus femoris [RF]) (Figure 1a). The proton density fat fraction (FF) was calculated by combining the water- and fat-only images obtained from the Dixon acquisition (Figure 1d). The lean volume and lean CSA were calculated by multiplying volume and global CSA, previously defined, by the water fraction (100%-FF).

FIGURE 1 |.

Representative qMRI images for a 73-year-old male subject. (a,e) Muscle segmentation from the Dixon water scan. (b) Fat fraction (FF), (c) axial diffusivity (AD), (d) fractional anisotropy (FA), (f) radial diffusivity (RD), and (g) T₂.

DTI data were processed using an in-house python-based pipeline. Diffusion volumes were denoised using patch2self [40] and non-rigidly registered to the Dixon water images. The diffusion-weighted images were fit to a tensor model using a weighted linear least squares (WLLS) algorithm. The tensor was then diagonalized to derive the diffusion eigenvectors (AD = λ₁, λ₂, and λ₃), from which RD (RD = (λ₂ + λ₃) / 2) and FA were derived (Figure 1b,c,e).

DTI-derived eigenvectors were utilized for deterministic fiber tracking. Seed points were uniformly distributed in each muscle ROI after erosion. Fiber tracking was performed with a step size of 3 mm, and tracts were terminated at the muscle borders. High tract density [37] or high fat percentage were used as stopping criteria. Additional stopping criteria were FA < 0.1, FA > 0.6, and fiber length > 150 mm. Fascicle length was calculated as the average length of each reconstructed tract within the muscle.

T₂ maps of lean muscle tissue (water T₂ of muscle mass, indicated from now on simply as T₂) were obtained from the MESE dataset using a non-linear least square fitting procedure based on extended phase graph (EPG) formalism using a two-component model (water + fat) to correct for fat infiltration and slice profile [41] (Figure 1f).

2.4 |. Skeletal Muscle Strength

Isometric, eccentric, and concentric strength were evaluated during a knee-extension exercise using an isokinetic dynamometer (HumacNorm). Both left and right legs were evaluated, and the order in which they were tested was randomized. All results were corrected for the effect of gravity. For all measurements, we used a standardized script to explain the procedures and motivate the subjects to apply their maximum torque. All subjects were allowed to perform the task twice at submaximal effort (once at 50% and once at 75%) prior to performing the task at maximum effort. Each participant performed five isometric contractions at maximum effort, as well as five isokinetic concentric contractions (at 120° and 90°/s) and five eccentric contractions (at 60° and 45°/s). Subjects were allowed to rest as needed in between different tasks. The total duration of the strength assessment procedures for both legs, including preparation, calibration, detailed explanation of the procedures, practice sessions at submaximal efforts, and resting, was approximately 45 min. The effective testing time for each leg at maximal effort was less than 90 s per leg, and the subjects were allowed to rest for at least 1 h between the strength testing session and the MRI scan. Every individual was able to complete all components of the assessment for both legs. Concentric, eccentric, and isometric torque values were determined as the highest torque value obtained during the five trials.

2.5 |. Statistical Analysis

An initial explorative analysis examined the association between muscle quantity (global volume and global CSA), torque, torque normalized by body weight, and age using a linear regression model. Spearman rank correlations were calculated between all imaging, torque, and demographic variables. All p-values were Sidak-adjusted for multiple comparisons. Univariate assessment of the effects of variables on torque was tested by linear regression adjusted for clustering between subjects. Multivariable assessment of the effects of variables on torque was performed by a mixed-effect regression, with random intercepts of subjects and an unstructured covariance matrix. Pseudo-R² was calculated as the square of the correlation between actual and predicted values. A p-value < 0.05 was considered significant.

The predictive power of the different torque models was tested using a repeated fivefold cross-validation analysis. This indicates how well a certain prediction model can generalize to an independent dataset.

3 |. Results

3.1 |. Participants Characteristics

We recruited 24 participants (12 female, range 30–79 years old, BMI = 25 ± 5 kg/mm², range 18.6–32.0 kg/mm²) between April 2022 and June 2023. Participant characteristics are summarized in Table 1. Global muscle volume and global CSA were strongly correlated (r = 0.98, p < 0.0001). Therefore, only global CSA was used as a measure of muscle quantity in subsequent data analysis. One subject had a missing T₂ measurement, and another one had missing FF data due to scanner issues.

TABLE 1 |.

Participant characteristics. Mean (std) of demographics, torque, and imaging parameters.

	Male (n = 12)	Female (n = 12)
Age [years]	58 (15)	58 (15)
Weight [kg]	79.8 (16.5)	68.0 (10.5)
Quadriceps CSA [cm2]	93 (16)	64 (13)
Volume [cm3]	1878 (404)	1194 (268)
Eccentric torque [Nm]	169.6 (44.1)	130.4 (42.3)
Concentric torque [Nm]	121.4 (36.0)	82.9 (27.2)
Isometric torque [Nm]	189.8 (43.7)	125.7 (37.7)
FA [-]	0.17 (0.02)	0.16 (0.02)
T2 [ms]	26.5 (2.5)	27.5 (2.2)
RD [mm2/s]	0.0018 (0.0001)	0.0017 (0.0001)
AD [mm2/s]	0.0022 (0.0001)	0.0022 (0.0001)
Fascicle length RF [mm]	64 (14)	55 (13)
Fascicle length VL [mm]	59 (11)	57 (11)

3.2 |. Skeletal Muscle Strength and Age

Torque was higher in males compared with females (154.4 ± 49.3 Nm vs. 110.5 ± 42.3 Nm, p < 0.001), even after normalizing by body weight (1.95 ± 0.52 Nm/kg vs. 1.62 ± 0.57 Nm/kg, p < 0.001).

There were no differences in torque between left and right leg (136.1 ± 52.2 Nm vs. 128.7 ± 49.4 Nm, p = 0.26).

Eccentric torque values were not different from isometric torque values (p = 0.26), while concentric torque was significantly lower (p < 0.001). To simplify analysis, torque was categorized as eccentric (velocity < 0°/s), concentric (velocity > 0°/s), or isometric (velocity = 0°/s) for further analysis.

Global CSA had a weak negative correlation with age (r = −0.433, p < 0.001). FF and T₂ showed a positive correlation with age (r = 0.648 and r = 0.457 for FF and T₂, respectively, p < 0.001), while shorter VL fascicle length was associated with increased age (r = −0.254, p < 0.001). All correlations between age and qMRI biomarkers are presented in Table 2. When isometric and isokinetic trials were analyzed separately, we observed that concentric muscle torque decreased with increasing age (r = −0.471, p < 0.001), but eccentric and isometric did not (r = −0.325, p = 0.080 and r = −0.416, p = 0.193, respectively). Stronger negative correlations with age were observed when the torque was normalized by body weight (r = −0.592, −0.451, and −0.549 for concentric, eccentric, and isometric torque, respectively, p < 0.001) (Figure 2).

TABLE 2 |.

Spearman correlations coefficients between torque, age, and qMRI biomarkers (cross-sectional area [CSA], fat fraction [FF], axial diffusivity [AD], radial diffusivity [RD], fractional anisotropy [FA], muscle T2, vastus lateralis fascicle length [Length VL], and rectus femoris fascicle length [Length RF]). All p-values are Sidak-corrected for multiple comparisons.

	Torque	Age	CSA	FF	AD	RD	FA	T2	Length VL	Length RF
Torque	—
Age	−0.338***	—
CSA	0.699***	−0.433***	—
FF	−0.429***	0.648***	−0.546***	—
AD	0.232*	0.078	0.191	−0.015	—
RD	0.254**	−0.016	0.169	−0.104	0.85***	—
FA	0.026	0.167	0.193	0.114	0.08	−0.24**	—
T2	−0.269**	0.457***	−0.513***	0.397***	0.33***	0.31***	−0.08	—
Length VL	0.241**	−0.254**	0.211	−0.283**	0.07	−0.02	−0.32***	−0.05	—
Length RF	0.299***	−0.207	0.325***	−0.179	0.16	0.10	−0.22*	0.07	0.73***	—

^*p < 0.05.

^**p < 0.01.

^***p < 0.001.

FIGURE 2 |.

Torque vs. age. Isometric, concentric, and eccentric torque, non-normalized (top) and normalized by body weight (bottom) vs. age. r: Spearman correlation coefficient. p-values are Sidak corrected for multiple comparisons.

Concentric and isometric torque had no correlation with BMI (p > 0.165) and a moderate positive correlation with eccentric torque (r = 0.439, p < 0.001). BMI was not correlated to muscle composition parameters (p > 0.9 for FF and T₂).

FF and T₂ had a positive correlation with age (r = 0.684 and 0.457, respectively, p < 0.001), while muscle fascicles in the VL and RF were shorter with increasing age (r = −0.25 and −0.21, p < 0.05) (Figure 3).

FIGURE 3 |.

qMRI parameters vs. age. Top: Global CSA, FF, T₂, and FA. Bottom: AD, RF, RF fiber length, and VL fiber length. r: Spearman correlation coefficient. p-values are Sidak corrected for multiple comparisons.

3.3 |. Univariable Model

The results of the univariable prediction of torque (when pooling results from all the isometric and isokinetic trials) are presented in Table 3. The stronger predictors of torque were muscle global CSA (R² = 0.453, p < 0.001) and lean CSA (R² = 0.459, p < 0.001), followed by FF (R² = 0.165, p < 0.001), RF fascicle length (R² = 0.086, p = 0.016), T₂ (R² = 0.054, p = 0.011), RD (R² = 0.044, p = 0.037), and AD (R² = 0.045, p = 0.013). FA was not an independent predictor of muscle strength (p = 0.680).

TABLE 3 |.

Univariable predictors of quadriceps torque. Significant correlations (p < 0.05) are indicated in italic font.

	R2	p
Lean CSA	0.459	< 0.001
CSA	0.453	< 0.001
FF	0.165	0.001
Length RF	0.086	0.016
T2	0.054	0.011
RD	0.044	0.037
AD	0.045	0.013
Length VL	0.063	0.084
FA	0.005	0.680

3.4 |. Multivariable Strength Models

Global CSA was strongly associated with all torque values (R² = 0.6521 for concentric, R² = 0.5276 for eccentric, and R² = 0.7983 for isometric, p < 0.001 for all). Only minor increases in R² were observed when considering the lean CSA instead of global CSA (R² = 0.6685 for concentric, R² = 0.5283 for eccentric, and R² = 0.8047 for isometric, p < 0.001 for all). FF was an independent predictor of torque, but the association was lost when correcting for global CSA (p > 0.275). The inclusion of RF fascicle length information improved the prediction of torque over global CSA alone for concentric and eccentric torque (R² = 0.6810, p = 0.010 and R² = 0.5490, p = 0.001, respectively) but not for isometric (p = 0.133). Similar results were observed for VL fascicle length (R² = 0.6843, p = 0.004 for concentric, R² = 0.5557, p = 0.017, and R² = 0.7980, p = 0.008 for isometric). RD had a positive correlation with eccentric torque after correcting for muscle global CSA (R² = 0.5561, p = 0.009). Adding FA to muscle quantity resulted in the best prediction model for concentric and eccentric torque (R² = 0.7129, p = 0.001 and R² = 0.5537, p = 0.002, respectively, compared with R² = 0.6521 and R² = 0.5276 for CSA alone). The contribution of FA to isometric strength after adjusting for CSA almost reached significance (R² = 0.8037, p = 0.051). The results are summarized in Table 4. The addition of T₂ proved to be significant for the prediction of eccentric torque (R² = 0.5478, p = 0.041). Age and sex were not significant covariates in the models (p > 0.05), nor was the IPAQ physical activity score (p > 0.07).

TABLE 4 |.

Quadriceps torque prediction models including qMRI parameters. p-values are indicated in italic font.

	Concentric	Eccentric	Isometric
Global CSA
CSA	< 0.001	< 0.001	< 0.001
R 2	0.6521	0.5276	0.7983
Lean CSA
CSA	< 0.001	< 0.001	< 0.001
R2	0.6685	0.5283	0.8047
Global CSA + FF
CSA	< 0.001	< 0.001	< 0.001
FF	0.275	0.888	0.639
R2	—	—	—
Global CSA + FA
CSA	< 0.001	< 0.001	< 0.001
FA	0.001	0.002	0.051
R2	0.7129	0.5537	—
Global CSA + T2
CSA	< 0.001	< 0.001	< 0.001
T2	0.097	0.041	0.139
R2	—	0.5478	—
Global CSA + RD
CSA	< 0.001	< 0.001	< 0.001
RD	0.075	0.009	0.961
R2	—	0.5561	—
Global CSA + AD
CSA	< 0.001	< 0.001	< 0.001
AD	0.438	0.127	0.763
R2	—	—	—
Global CSA + Length RF
CSA	< 0.001	< 0.001	< 0.001
Length RF	0.010	0.002	0.133
R2	0.6810	0.5490	—
Global CSA + Length VL
CSA	< 0.001	< 0.001	< 0.001
Length VL	0.004	0.017	0.008
R2	0.6843	0.5557	0.7980

3.5 |. Cross Validation

The results for the cross-validation prediction models are displayed for torque in Supporting Information. The same models presented in Table 4 are included in Table S1. Only the models where the inclusion of qMRI parameters improved the prediction of torque are included. The results from the cross-validation analysis were similar to the step-wise regression analysis and showed improved prediction of knee extension torque over muscle global CSA when adding FA, RD, T₂, and muscle fascicle length values.

4 |. Discussion

This study tested the hypothesis that, after controlling for muscle quantity (global CSA), qMRI parameters in the quadriceps could improve the prediction of knee extension torque. Our results show that the FA, T₂, and muscle fascicle length in the quadriceps were significantly associated with knee extension torque after adjusting for muscle CSA, thus supporting our hypothesis.

Many cross-sectional and longitudinal studies, including ours, have reported age-related changes in global muscle CSA that are associated with a decline in muscle strength [42]. Muscle quantity and strength are closely related, but their relationship is not strictly linear, as evidenced in contexts such as bodybuilding, where increases in lean muscle mass do not always result in proportional gains in strength. In bodybuilding, where training often emphasizes muscle size, the focus is on hypertrophy rather than maximal force output. Consequently, individuals may gain substantial muscle mass without equivalent improvements in strength. This discrepancy can be attributed to several factors, including neural adaptations, which can occur independently of hypertrophy. Additionally, muscle architecture and fiber type composition also influence force production, highlighting the importance of considering muscle quality in addition to muscle quantity.

A non-linear relationship between muscle quantity and strength has also been observed in the context of muscle aging, where strength declines much faster than muscle mass. In fact, while the annual decrease in muscle mass after the age of 70 amounts to about 1%, muscle strength declines approximately three times faster [43]. Additionally, maintaining or gaining muscle mass does not prevent age-associated declines in muscle strength [43], suggesting that functional deficits in older adults are caused by the decline in both muscle quantity and quality. Muscle aging and sarcopenia are complex and multifactorial conditions, characterized by changes in muscle architecture and muscle composition, such as fatty infiltration and selective type 2 fiber atrophy. Our study suggests that these changes in muscle quality can be non-invasively assessed using chemical shift MRI, DTI, and T₂ relaxometry, and that they contribute to overall muscle strength independently of global muscle CSA. This aligns with a recent US study showing that the inclusion of echogenicity, a non-specific marker of muscle composition, can predict strength independently of muscle mass [44].

Beyond changes in global CSA with age, we also detected a dependence of qMRI parameters on age. Specifically, in our cohort, FF increased by approximately 1% for every decade of age, consistent with previous studies [45, 46], reflecting increasing fat infiltration. However, this relationship is likely non-linear across all decades, as fat infiltration is known to accelerate in older age groups [45]. The FF values in this study, between 2% and 12%, fall within literature values for healthy muscles [47], but a positive correlation has been reported between frailty index and FF, suggesting the presence of elevated FF in pre-frail and sarcopenic subjects [48]. In univariable analysis, this increased FF was negatively correlated to torque, as also reported by Baum et al. [49], likely indicating a reduction in the amount of contractile tissue [32]. Interestingly, although FF was a strong predictor of muscle torque when considered in isolation, it was not significant after adjusting for global muscle CSA. While FF has been identified as a strong independent predictor of muscle strength in disease [50], our study shows that this relationship might not hold in healthy subjects. This is likely due to the fact that our cohort only included non-sarcopenic and non-frail subjects with relatively low FF, as also noted by Smith et al. [51]. The T₂ values in this study were in the range 24–34 ms, as reported by Keene et al. [52] for healthy subjects. We also observed an age-related increase in muscle T₂, consistent with reports for both the thigh [53] and the calf [47]. However, unlike Yoon et al. [54], we did not observe a correlation between quadriceps T₂ and knee extension torque for isometric contraction. This is likely explained by the fact that by utilizing a multicompartment model for T₂ calculation, we could isolate the T₂ of lean muscle tissue and are therefore less sensitive to the presence of fatty replacement, which is known to negatively affect muscle strength [32] and function [55].

Cameron et al. analyzed a larger cohort and found a small positive correlation between FA and age, and a small negative correlation between MD and age [56]. In contrast, we did not observe any correlation between FA and age, in agreement with Yoon et al. [54]. We focused on RD rather than MD, as RD more closely reflects restriction in the cross-section of muscle fibers, and can potentially be a better marker for muscle fiber atrophy [57]. These associations have been interpreted in terms of sensitivity of DTI to muscle fiber size and microstructure. Under these assumptions, a positive correlation between RD and muscle strength is expected, as fiber loss and cellular atrophy are primary contributors to weakness in sarcopenia and aging [17]. A positive association between RD and muscle strength has been identified in several studies [35, 58], but was not directly tested by Cameron et al. [56]. Despite these correlations between DTI parameters and muscle strength, DTI alone is unlikely to be a strong predictor of muscle health [59]. Furthermore, as smaller muscle fibers can decrease overall muscle volume, changes in RD might simply reflect global muscle size. A key difference between the results of previous studies and the present one is the combined investigation of MRI biomarkers of muscle composition and architecture, and their association with strength, independent of muscle size. The positive correlation observed in this study between knee extension torque and RD, even after adjusting for global muscle size, further supports the role of RD as a non-invasive biomarker of cellular size, and confirms its potential in sarcopenia and aging research.

In our study, FA was not correlated to muscle torque nor age, but became a significant predictor of torque after adjusting for muscle CSA. This observation, along with the significant role of T₂, suggests the sensitivity of qMRI to muscle fiber type. Age-related increases in T₂ have been linked to the disproportionate reduction in size of type 2 fibers compared with type 1 fibers, which is believed to be a key mechanism leading to weakness in older adults [17]. As this process leads to an increased volume proportion of type 1 fibers, which are characterized by higher T₂ values [17], the selective type 2 fiber atrophy would explain the positive correlation with age observed in this study and the significant independent effect of T₂ in the prediction of muscle strength. The contributions of selective type 2 fiber atrophy to the reduced muscle strength are further strengthened by the significant correlation between strength and FA, which has been associated with immunochemistry-derived [36] in younger male subjects. Our results suggest that the findings may extend to female and older subjects, making DTI a promising tool to estimate the proportion of type 2 fibers. However, direct comparison between these qMRI parameters and histology is needed to confirm the biological underpinning of our findings.

We measured a shortening in muscle fascicle length with age [20], which was associated with reduced force production, as previously reported [20]. This shortening can be explained by a loss of sarcomeres in series [60], and is more pronounced in frail individuals [61]. The 3D assessment of quadriceps architecture offers valuable insight into muscle adaptations with aging, and mechanisms of strength loss in older adults.

All the subjects in this study were considered healthy and physically active. Wider variability in qMRI parameters is expected in (sarcopenic) older adults, together with stronger correlations with muscle strength.

For this study, the quadriceps were manually segmented. However, tools are currently available for accurate and automatic muscle segmentation [51, 62, 63]. These tools hold great potential to streamline the extraction of qMRI parameters and facilitate their adaptation in clinical and research settings.

4.1 |. Limitations

The diffusion-weighted scans in this work were collected along a limited number of non-collinear diffusion-encoding directions, which could negatively impact the accuracy of the diffusion tensor fit. This was required to obtain a tradeoff between high angular resolution and sufficiently high SNR. However, considering the relatively simple geometry of the quadriceps muscle, 15 directions are likely enough to capture the main architectural features of muscle fibers [64].

One of the limitations of the current study is its cross-sectional nature, which did not allow confirming whether changes in muscle quantity and quality are predictive of future changes in muscle strength. Additionally, our cohort only included 24 healthy subjects, so the conclusions of this study might not directly apply to frail and sarcopenic older adults. However, due to the worse muscle quality in sarcopenic subjects, we expect an even higher percentage of variance in strength production to be explained by muscle quality in sarcopenic adults compared with healthy older adults. Furthermore, this study defines age-related normative values for qMRI biomarkers of muscle quality that could be used in the future to differentiate sarcopenia from “healthy aging.” Our analysis revealed no correlation between BMI and muscle composition metrics and a positive correlation between eccentric torque and BMI. However, our study only included mild obesity, so the correlation between BMI and muscle strength might not generalize to subjects with larger BMI, in particular, subjects with sarcopenic obesity, where a high BMI is associated with low muscle mass. While further research will be needed to fully characterize muscle quality in sarcopenic obesity and its correlation with strength, our results suggest the limited value of BMI for skeletal muscle characterization.

Future studies should further investigate the connection between qMRI parameters and muscle function in older subjects at risk of developing sarcopenia, such as frail and pre-frail individuals.

5 |. Conclusions

This work highlights the connection between qMRI parameters and muscle strength. While a complete and definitive explanation of the biophysical underpinning of qMRI and parameters in skeletal muscle is still lacking, we have shown the combined effect of muscle quantity (CSA) and quality (FA, RD, T₂, and fiber length) on knee extensor torque. Based on these observations, qMRI holds promise for the objective and non-invasive evaluation of skeletal muscle in older adults. The combined evaluation of muscle quality and quantity presented in this study can lead to earlier and more accurate assessment of sarcopenia in older adults, and provide specific targets for treatments in the future.

Supplementary Material

Supporting Information

Additional supporting information can be found online in the Supporting Information section. Table S1: Quadriceps torque prediction models including qMRI parameters with 5-fold cross-validation. Only the models based on CSA or where the inclusion of qMRI parameters improved the prediction of torque compared to CSA alone are included. Table S2: Quadriceps torque prediction models including lean CSA and qMRI parameters.

Abbreviations:

AD

axial diffusivity

CSA

cross-sectional area

EWGSOP

European Working Group on Sarcopenia in Older People

FA

fractional anisotropy

FF

fat fraction

MESE

multi-echo spin echo

RD

radial diffusivity

RF

rectus femoris

VI

vastus intermedius

VL

vastus lateralis

VM

vastus medialis

WLLS

weighted linear least squares

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.