Short- and long-term reproducibility of diffusion-weighted magnetic resonance imaging of lower extremity musculature in asymptomatic individuals and a comparison to individuals with spinal cord injury

Diffusion-weighted magnetic resonance imaging (DW-MRI) provides a potential diagnostic and prognostic measure for the normal (and abnormal) movement of water through living soft-tissues. Skeletal muscle is an emerging application for DW-MRI [1], where it holds considerable potential in common, yet enigmatic, clinical conditions. For example, significant changes in DW-MRI metrics have been observed in individuals with rotator cuff tears [2], muscular dystrophy [3], myopathies [4, 5], radiculopathy [6, 7], and in animal models of denervation [7, 8]. In cases, diffusion-related changes may also precede findings on other diagnostic tests [7], opening a critical early window where diagnostic and prognostic data could better inform intervention. Additionally, DW-MRI could serve as a marker of muscle response to therapeutic exercise [9, 10]. Despite the increasing number of investigations utilizing DW-MRI in skeletal muscle – and in particular the use of skeletal muscle DW-MRI as a diagnostic or outcome measure – normative data remains sparse. Complicating matters further, the rater-based reliability and temporal reproducibility of DW-MRI-related parameters have also yet to be established for skeletal muscle.

Here, we investigate a clinically relevant DW-MRI parameter, the Apparent Diffusion Coefficient (ADC), in muscles of the lower limb. We estimate the rater-based reliability and the temporal reproducibility of ADC at both short- and long-inter-scan intervals, with the primary goal of providing an additional statistical reference for future investigations incorporating DW-MRI of skeletal muscle. We also quantify the ADC both in a cohort of individuals without neurological or musculoskeletal injury and in a cohort of individuals with motor incomplete spinal cord injury, a potential target for clinical use of DW-MRI. We hypothesize that ADC will be elevated in leg muscles innervated by spinal nerves below the level of lesion, given the previously documented elevations in ADC following denervation and/or trauma [2, 7].

The objectives of this investigation were to provide normative ADC values in healthy skeletal muscle at 3 T, to further establish the rater-based reliability and temporal reproducibility of DW-MRI in muscle, and to estimate ADC in a cohort of individuals with motor incomplete spinal cord injury. Although previous reports of mean ADC for healthy skeletal muscles exhibit considerable variability, our estimates are generally consistent with the available data. In the earliest results, obtained at 0.5 T, Morvan et al. reported mean ADC of resting muscle at 1.82 × 10^− 3 mm²/s [14]. ADC estimates obtained at 1.5 T have generally been lower than our results, ranging from 1.26 × 10^− 3 mm²/s to 1.51 × 10^− 3 mm²/s [7, 15, 16], although more recently, values of 1.8 × 10^− 3 mm²/s have again been reported (albeit at the shoulder) [2]. ADC estimates of leg muscles obtained at 3 T appear to be more closely aligned with our findings, ranging from 1.51–1.85 × 10^− 3 mm²/s [3‐5].

Overall, we found excellent absolute rater-based reliability and short-term reproducibility for ADC in skeletal muscle. Our average rater-based LSC was less than 10% across muscles, and our average LSC for short-term reproducibility was 4.5% for scans acquired on the same day. Both findings are consistent with the available literature [17, 18]. Based upon standard reporting conventions for relative reliability metrics [19], our ICC estimates for rater-based reliability and short-term reproducibility were also excellent. We caution against direct comparisons of our ICC values to other studies however, given that ICC estimates are highly sensitive to sample size and across-participant variability not attributable to the instrument itself. For example, while Heemskerk et al. [18] report much lower ICCs of 0.7 and 0.17 for rater-based reliability and short-term reproducibility, their repeatability coefficients (i.e., LSC) for the same data were closely aligned with our results, at 6 and 12%, respectively.

Experimental and/or physiological factors likely contributed to our lower long-term ICC estimates. For example, our study implicitly assumed that ADC values in healthy skeletal muscle are stable over time in the absence of trauma, disease, etc. This is a clear oversimplification, as changes in temperature, activity level, and hydration are all capable of influencing the ADC of skeletal muscles [9, 10, 15, 20]. Because we did not explicitly control for these factors, our slightly lower long-term performance is presumably related, at least in part, to true changes in muscle physiology that were unaccounted for by our experimental design. Nevertheless, our estimate of absolute long-term reproducibility is consistent with those reported in the literature, which range from repeatability coefficients of 7–15% with scans acquired 1–2 weeks apart [18, 21].

We found a modest but statistically significant increase in average ADC across muscles in individuals with motor incomplete SCI compared to individuals without neurological or musculoskeletal injury. These increases were confined to the plantarflexor muscles (MG, LG, and SOL), with the TA showing no clear differences between cohorts. Given that elevations of muscle ADC, not reductions, are most commonly associated with pathology, it was somewhat surprising that average ADC was higher in individuals with SCI for whom walking is the primary means of ambulation compared to wheelchair users. It should be reiterated however that neural drive to leg muscles was not altogether absent in our wheelchair user cohort; these participants could perform standing transfers and engage in some therapist assisted walking. Nevertheless, changes in ADC both within the SCI cohort and between the SCI and control cohorts presumably reflect a combination of direct, SCI-induced alterations in neural drive to muscles and indirect effects associated with reduced motor output. Indeed, denervation, atrophy, and changes in physical activity level are all known to influence skeletal muscle ADC [7, 8, 10, 22].

Because our study was not intended to correlate ADC with motor function, the relevance of elevated ADC to impairment – if any – remains unclear. Nevertheless, it is interesting to note that plantarflexor spasms represent one of the most common presentations of spasticity after SCI [23], and we observed preferential elevation of ADC in these muscles. Future work that explores the mechanisms of ADC elevation post-SCI and their potential relevance to motor function, as well as normative estimates of ADC in motor complete SCI, is clearly warranted.

Our study has important limitations. Unlike previous investigations, our ROI definition was performed by raters with minimal imaging experience yet an intimate knowledge of the relevant anatomical structures (Doctor of Physical Therapy students). Although this choice was motivated by a desire to estimate a potential lower bound on rater-based reliability, it also complicates the interpretation of our findings by making it harder to decouple the sources of experimental, physiological, and rater-induced variability. Another limitation is that we used the same MRI scanner for all imaging sessions. As a result, our ADC values, as well as our reliability estimates, should be interpreted only in this context. In purest terms, our results may over-estimate rater-based reliability and/or reproducibility and thus be much different if compared to results from multi-site investigations incorporating different community imaging centers. Finally, we did not control leg temperature in the cohort of individuals with SCI. Thus, it is possible that subtle temperature differences existed between the SCI and control cohorts and/or between the more and less impaired legs of individuals with SCI. Reduced temperature in the extremities, as has been reported in SCI [24], could change our estimates of ADC independently of SCI-induced alterations of muscle physiology [15, 20]. It seems unlikely that changes in temperature could explain the differences in ADC we found between individuals with and without SCI, however. Specifically, leg temperature in individuals with SCI is generally lower than that of individuals without SCI [24], suggesting that ADC should be lower in SCI than controls [20]. However, we found a significantly elevated ADC across muscles in the SCI cohort compared to the control cohort, indicating that temperature was not the primary factor driving the observed ADC differences. We also found no significant differences in ADC between legs within the SCI cohort for any muscle (Fig. 2). This finding also suggests that the influence of leg temperature on our results was likely minimal (although it was presumably a source of overall variability in our estimates of ADC). Nevertheless, future work investigating skeletal muscle ADC in individuals with SCI (or other neurological injuries) should systematically monitor and control extremity temperature to avoid undue experimental confounds.

Despite the promise of DW-MRI in skeletal muscle as an additional tool for clinical evaluation and basic scientific investigation, its utility has remained limited at least in part due to a lack of normative data and an incomplete assessment of its precision and reliability. Our results demonstrate that its reproducibility is excellent over short time scales, reinforcing the available literature. We also extend these findings by providing insights into the stability of the approach over the longest inter-scan interval available to date (9 months). Although performance was expectedly lower over this time scale, ADC changes as small as 12% were still discernable. Considering that ADC changes of ~ 24% have been reported in radiculopathy [7], ~ 35% following rotator cuff tear [2], and ~ 20% immediately following vigorous contraction of the same muscles included in this investigation [10], we conclude that controlled experiments using DW-MRI of skeletal muscle are sufficiently reproducible for longitudinal analyses. With evolving scanner and sequence technologies, DW-MRI may prove to be a highly sensitive measure of physiological changes to peripheral muscles after neurological injury and in any number of common, yet enigmatic, neuro-musculoskeletal conditions (e.g., low back pain, whiplash). Such knowledge could help to improve clinical decision making at a time when therapeutic intervention may have its largest effect.