Introduction

The blood oxygen dependent level (BOLD) effect uses deoxy/oxy-hemoglobin dependant changes in local magnetic susceptibility to infer changes in blood-flow and metabolic need.¹ Most noted as the mechanism underlying brain fMRI, it also has shown utility in characterizing skeletal muscle physiology in healthy and diseased populations.² However, for diagnosis of neuromuscular disorders muscle biopsy is the gold-standard.³ Muscle biopsy identifies the ratio of type I (high-oxidative/slow-twitch(ST)) to type II (low-oxidative/fast-twitch(FT)) muscle fibers aiding in diagnosis. Electromyography, also used clinically, does not provide fibre specific information.³ A non-invasive technique to characterize skeletal muscle fibre composition, would be clinically invaluable. The hypothesized microvasculature signal origin and differing perfusion demands of type I and II fibres suggest that the muscle BOLD signal could non-invasively characterize muscle fibre profiles.^3,4 The temporal fractal dimension (FD), a measure of signal complexity, is proposed to provide this insight.

Methods

Eight male subjects (4 endurance (ST): 4 power (FT) athletes) were tested for their calf raise one repetition maximum (1-RM) using the Epley estimate.⁵ Predominate fibre type groupings were confirmed based on number of 80% 1-RM repetitions.⁶ Prior to resting-state lower leg functional image collection, participants lay in the scanner bed for 30min to allow for normalization of blood-flow in the legs, shown to affect the skeletal muscle BOLD signal.⁷ A 16-channel transmit/receive extremity coil and GE 3T MR750 scanner were used. Proton density weighted, fat-supressed images (0.625x0.625x4mm, 15 slices (1mm spacing), TE/TR/flip=30/3000ms/111⁰) provided anatomic reference. Using the FMRIB Software Library, functional images (2.5x2.5x10mm, 2 slices (no spacing), TE/TR/flip=35/109ms/70⁰, 2434 volumes) were motion corrected, and gastrocnemius (FT dense) and soleus (ST dense) muscles were manually segmented, using the anatomical reference, for their differing twitch fibre profiles.⁸ The first and last 193 volumes were truncated due to the short TR not satisfying the Ernst angle condition and motion considerations respectively.⁹ Time-series were normalized by dividing by the first time point, to account for coil sensitivity and proximity signal amplitude dependence. Voxel-wise BOLD timeseries, mono-FD and bi-FD were computed using the scaled windowed variance approach, with linear detrending, removing scanner induced low-frequency variations.¹⁰ The segment-wise FD for each voxel was computed over the four 512-point sections, then averaged across segments. Averaging four temporal data points improved voxel FD measure reliability, which is impacted by linear fitting slope estimates. Furthermore, since temporal variability appears invariable at large window sizes, FD measures would be unreliable without segmentation.¹¹ The mono-FD and bi-FD were computed by estimating the slope by fitting one line and two lines, respectively, to the non-saturated portion of the log-log plot for a given voxel and segment. Outlier detection of values greater than three standard deviations from the mean helped to exclude voxels containing connective tissue or vasculature. All mono-FD and bi-FD voxel distributions were negatively skewed. Post log transform correction, distributions failed Jarque-Bera normality tests, therefore Wilcoxon rank sum tests were used to examine within participant differences and Kruskal-Wallis to test within and between group differences.

Results

Participant comparison of FD between the gastrocnemius and soleus is summarized in figure 1. Notably the mono-FD, bi-FD component 1 (C₁) and 2 (C₂) all distinguished the difference between muscles for 5/8 participants. The mean participant bi-FD for C₁ and C₂ are detailed in figures 2 and 3. The gastrocnemius FD was less than that for the soleus, regardless of component, for 7/8 participants. For C₁ the four FT participants showed a lower complexity than the ST counterparts, but their distribution appeared random for C₂. The Kruskal-Wallis tests showed significant differences between FT and ST groups regardless of muscle or component with greater differences within/between groups using C₁ and between groups in the soleus.

Discussion

In both activity groupings, for both FD components, the gastrocnemius FD was lower than the soleus. Furthermore, in both muscles, for both FD components, the FD for the FT group was less than the ST. Both observations suggest a higher FD indicates more ST fibres. Additionally, the bi-FD components had variable discriminability. Specifically, the average p values for C₁ ST=0.0468, FT=0.0444, C₂ ST=0.5909, FT=0.0053 and average C₁ to C₂ improvement FT=96.43%, ST=-84.75%. Therefore, C₁ classifies both groups better than C₂ and the relative classification is worse for FT with C₁ and ST with C₂. Since we expect muscles to be more homogenous for the given fibre type grouping, a decrease in classification accuracy would suggest a sensitivity to fibre type. Additionally, the FT related component would be expected to have greater predictive power as there is a larger difference in FT content between muscle groups. Therefore, this suggests that C₁ is influenced by FT and C₂ by ST fibres.

Conclusion

Mean voxel bi-FD differentiated the soleus and gastrocnemius within participants and based on endurance/power groupings C₁ was shown to be influenced by FT and C₂ by ST fibres. This indicates that muscle BOLD bi-fractal characterization could non-invasively provide information on muscle fibre type ratios. Verification of ST and FT composition via muscle biopsy is required to validate these preliminary observations. Nonetheless, this work sets the foundation for development of a non-invasive tool to replace muscle biopsy to assess muscle fiber types, revolutionizing neuromuscular disorder diagnosis.

Acknowledgements

Thank you to the St. Joseph's MRI staff for aiding in the data collection process. Funding was provided through an Ontario Graduate Student scholarship granted to JEM and a NSERC Discovery Grant (RGPIN-2017-06318) to MDN.

References

1. Ogawa S, Lee TM, Kay AR, et al. Brain magnetic resonance imaging with contrast dependent on blood oxygenation Proc Natl Acad Sci. 1990;87(24):9868–9872.

2. Jacobi B, Bongartz G, Partovi S, et al. Skeletal muscle BOLD MRI: From underlying physiological concepts to its usefulness in clinical conditions. J Magn Reson Imaging. 2012;35:1253-1265.

3. Joyce NC, Oskarsson B, Jin LW. Muscle biopsy evaluation in neuromuscular disorders. Phys Med Rehabil Clin N Am. 2012;23(3):609-631.

4. Behnke BJ, Armstrong RB, Delp MD. Adrenergic control of vascular resistance varies in muscles composed of different fiber types: influence of the vascular endothelium. Am J Physiol Regul Integr Comp Physiol. 2011;301(3):R783-R790.

5. LeSuer DA, McCormick J, Mayhew J, et al. The Accuracy of Prediction Equations for Estimating 1-RM Performance in the Bench Press, Squat, and Deadlift, J Strength Cond Res. 1997;11(4):211-213

6. Hall ECR, Lysenko EA, Semenova EA, et al. Prediction of muscle fiber composition using multiple repetition testing. Biol Sport. 2021;38(2):277-283.

7. Elzibak AH, Noseworthy MD, Assessment of diffusion tensor imaging indices in calf muscles following postural change from standing to supine position. Magn Reson Mater Phys. Biol Med. 2014;27:387–395

8. Jenkinson M, Beckmann CF, Behrens TE, et al. FSL. Neuroimage. 2012;62(2):782-790.

9. Helms G, Dathe H, Weiskopf N, et al. Identification of signal bias in the variable flip angle method by linear display of the algebraic Ernst equation. Magn Reson Med. 2011;66(3):669-677.

10. Cannon MJ, Percival DB, Caccia DC, et al. Evaluating scaled windowed variance methods for estimating the Hurst coefficient of time series. Physica A. 1997;241(3-4):606-626.

11. Chung HW, Chung HJ. Correspondence re: J. W. Baish and R. K. Jain, Fractals and Cancer. Cancer Res., 60: 3683–3688, 2000. Cancer Res. 2001;61(22):8347-8348.