Test-retest reliability of muscle fiber conduction velocity and fractal dimension of surface EMG during isometric contractions

The study was approved by the local ethics committee of the Swiss Italian Health and Sociality Department, Switzerland. All procedures were conducted according to the Declaration of Helsinki. All participants signed a written informed consent form before participation in the experiments. Forty healthy recreationally active volunteers (20 women and 20 men) aged between 20 and 33 years from a university setting were recruited to participate in the study.

The subjects participated in two experimental sessions ('trial 1' and 'trial 2'), a week apart at the same time and under the same environmental conditions. Measurements were performed on the two arms randomly within each session.

Subjects were seated in a height-adjustable chair with their arm positioned on an isometric ergometer (MUC1, OTBioelettronica, Turin, Italy), equipped with a load cell (Model TF022, CCT Transducers, Turin, Italy). In order to isolate the action of the biceps brachii, the wrist was fastened to the ergometer, with the elbow at 120°, as shown in figure 1.

Zoom In Zoom Out Reset image size

Initially, two isometric MVCs of 2–3 s were performed, separated by 2 min rest. During each contraction, the force trace was displayed to participants on a computer monitor as visual feedback. Participants were instructed to increase the force up to their maximum, and to hold it as steady as possible. Participants were given verbal encouragement.

Next, after 2 min rest, a low-level contraction (20% MVC) was performed for 120 s, after which the subjects were asked to provide a value of the perceived exertion on a visual Borg scale, ranging from 6 to 20 (Borg 1982). After 5 min rest, the subjects were asked to perform a higher level contraction (60% MVC) to be maintained until exhaustion. During the contraction subjects were verbally encouraged to keep the force level for as long as possible, until the force value decreased below 90% of the target (endurance time, i.e. the time for which a subject is able to maintain the requested mechanical task (Merletti and Roy 1996)).

Myoelectric signals were detected from the biceps brachii, in a monopolar configuration using a bidimensional array of 64 electrodes (3 mm diameter, 8  ×  8 grid, 10 mm interelectrode distance (IED); model ELSCH064NM3; OT Bioelettronica) (figure 1). This muscle was chosen in order to obtain high-quality sEMG signals according to the qualitative criteria described in Beretta-Piccoli et al (2014). The electrode grid was applied on the muscle belly, with its distal edge close to the cubital fossa and the midline of the array aligned with the midline of biceps along a line from the cubital fossa to the acromion (see figure 1). A ground electrode was placed on the contralateral wrist. The EMG signals were amplified (EMG-USB2; OT Bioelettronica), band-pass filtered (10–750 Hz), sampled at 2048 Hz, and stored on a computer.

The isometric ergometer was used to measure elbow torque with a torque meter operating linearly in the range 0–1000 Nm. The torque signal was amplified (MISO II; OT Bioelettronica) and stored on a computer with the sEMG data. The torque signal was displayed on a screen, providing real-time biofeedback

The number of channels used for CV estimation was selected based on visual inspection, which comprised five steps:

• (1)  
  Identification of movement artifacts or missing channels: signals with large amplitude changes due to movement were removed from the analysis.
• (2)  
  Identification of power line interference: signals which presented large sinusoidal components at 50 Hz or their harmonics, were removed from the analysis.
• (3)  
  Manual identification of the main innervation zone (if present under the array) and of the distal tendons (through observation of propagating waves and inversion of MUAP shapes).
• (4)  
  Selection of the array column where the maximal amplitude and largest MUAPs were visible.
• (5)  
  Selection of the channels between the innervation zone and the distal tendons, where the MUAPs appear similar in shape and shifted in time.

The number of channels chosen to estimate CV was between 4 and 7, i.e. the maximal number of channels between the main IZ and the distal tendon. The minimal number of channels was set to 4 according to a previously published study on the reproducibility of CV estimation using multichannel sEMG (Farina et al 2004).

The FD was computed on each of the signals selected for CV computation and then averaged; thus, for each signal epoch, one value of CV and one value of FD were obtained.

FD was estimated using the box-counting method, as previously reported (Gitter and Czerniecki 1995). Briefly, a grid of square boxes is used to cover the signal, and the number of boxes that the sEMG waveform passes through is counted. When the box size decreases, the number of the boxes that are counted will increase exponentially. The range of box size is restricted in order to avoid saturation for high and low value of size (Gitter and Czerniecki 1995). The box size was fixed to 13 steps equally spaced in logarithmic scale, with the smallest box equal to 1/128th of a second and the largest box equal to 1/8th of a second. The vertical side of the boxes was normalized to the range of the signal during epochs of 1 s and divided in the same number of boxes.

However, by plotting the logarithm of the number of boxes required to cover the signal versus the logarithm of the inverse of the box area, the exponential relationship becomes approximately linear. The slope of the interpolation line (estimated using the least mean squared procedure) is the fractal dimension (FD) (Mesin et al 2009). Therefore, the following expression defines the FD:

$$\begin{eqnarray}&&\text{FD}=~\frac{\log N}{\log \frac{1}{L}}\end{eqnarray}$$

where N is the number of boxes required to cover the signal and L is the box side, with the ratio indicating the slope of the interpolation line.

Eventually, CV was estimated using a multichannel algorithm (Farina and Merletti 2003) on single differential signals, based on the matching between signals filtered in the temporal and in the spatial domains, using non-overlapping signal epochs of 1 s. CV values outside the physiological range (3–8 m s⁻¹) were excluded from the analysis.

Data were analyzed by a custom-written software in MATLAB R2014b (Mathworks, Natick, USA)

Linear regression over time was applied to FD and CV in order to extract an initial value and slope.

Statistical tests were conducted for FD and muscle fiber CV initial values and slopes, considering the data from left and right biceps brachii as a single set. A Shapiro-Wilk test revealed that all these variables were normally distributed. Data are reported as means  ±  standard deviation (SD) within the text and tables.

Test-retest reliability of the variables was examined using the intraclass correlation coefficient (ICC_(2,1)) (Weir 2005) and the Bland–Altman plot (Bland and Altman 1986) since their use has been recommended in reliability studies (Rankin and Stokes 1998, Bruton et al 2000, Weir 2005). The criteria used for the interpretation of the ICCs were as follows: 0.00–0.25: no correlation; 0.26–0.49: low correlation; 0.50–0.69: moderate correlation; 0.70–0.89: high correlation; 0.90–1.00: very high correlation (Munro 2005). Bland–Altman plots were provided to give a visual representation of the size and range of differences between two consecutives measurements from the same subjects (i.e. limits of agreement: average differences  ± 1.96^* standard deviation of the difference). The standard error of measurement (SEM), defined as the amount of error that might be considered as a measurement error, was calculated using the following equation (Stratford and Goldsmith 1997):

$$\begin{eqnarray}&&\text{SEM}=\sqrt{\sigma _{\text{T}}^{2}\left(1-\rho \right)}\end{eqnarray}$$

where $\sigma _{\text{T}}^{2}$ represents the total variance and $\rho$ the ICC.

Moreover the minimal detectable change (MDC₉₅) for the ICC_(2,1) was also calculated to provide an estimate of the true change for each variable for a participant using the following equation (Weir 2005):

$$\begin{eqnarray}&&\text{MD}{{\text{C}}_{95}}=1.96\times \text{SEM}\times \sqrt{2}\end{eqnarray}$$

The MDC₉₅ is the minimum amount of change in subject's variables that ensures the change is not the results of measurement error: therefore, values greater than the MDC₉₅ represent true change.

Statistical analyses were performed using SPSS Version 22.0 (SPSS Inc, Chicago, IL, USA), and significance was set to α  =  0.05.

The issue of reliability of CV initial values and slope is a complex issue that has been addressed by several studies in the past. Farina et al (2004) suggested that there are at least three factors affecting CV reproducibility: (1) variance of estimation of CV intrinsic in the method, (2) electrode repositioning and (3) variability intrinsic in the subjects' performance and modality in muscle control. To reduce the impact of factors (1) and (2), we used bi-dimensional arrays with 10 mm IED, in order to have larger number of electrodes, optimal distance between detection points and lower sensitivity to electrode displacements, thus reducing the experimental noise.

Initial values of CV showed high to moderate relative test-retest reliability during the low and high level contractions (table 2), and the same behavior emerged from the absolute test-retest reliability analysis depicted by the Bland & Altman plot (figures 3(a) and (b)) and by the low values of SEM (0.045–0.051 m s⁻¹), which indicates the expected trial-to-trial noise. Therefore, initial value of CV, estimated with multichannel sEMG, is a reproducible variable, suitable for example to characterize MU types constituency of muscles (Sadoyama et al 1988, Farina et al 2004), for differentiating pathological conditions (e.g. Klaver-Król et al 2012, Butugan et al 2014, Boccia et al 2016b) or to highlight fiber size differences (Blijham et al 2006).

Despite these considerations, a negative ICC of CV slope was found at 20% MVC (table 2), indicating a very low inter-subject variability. Given the fact that CV slopes are indicative of the development of peripheral muscle fatigue (Merletti and Parker 2004), and that at 20% MVC, as previously reported (Beretta-Piccoli et al 2015), peripheral fatigue is almost absent across the subjects, the inter-subject variance will be very low, so that the relative test-retest reliability expressed through the ICC may become negative. Moreover, the values distribution in the Bland & Altman plot (figure 3(c)) results in a very narrow cloud around zero, with many values overlapped, thus confirming a low inter-subject variability. Additionally, MDC₉₅, which provides an estimate of the magnitude of change on CV slope and, therefore, accounts for measurement error and variability (Weir 2005), and SEM are both very low, respectively 0.025 and 0.009% s⁻¹, thus supporting the fact that the low test-retest reliability is not due to a measurement issue.

On the contrary, the fact that the 60% MVC contraction was much more fatiguing (CV slope was negative), likely lead to a higher inter-subject variability, as depicted in the Bland & Altman plot (figure 3(d)), together with a slight increase of SEM and MDC₉₅.

In literature there are few studies addressing the issue of CV slope test-retest reliability during isometric contractions at 50 or 60% MVC, using the ICC (Rainoldi et al 1999, Rainoldi et al 2001, Falla et al 2002, Farina et al 2004, Ollivier et al 2005) and their conclusions are rather consistent indicating that ICC ranges from low to moderate. In most of these investigations, a four-electrodes system to estimate CV (with an IED of 5 mm) was used: the sensitivity to electrode location, was identified as the main factor limiting test-retest reliability. In fact, electrode repositioning between sessions produced high levels of experimental noise. Our results confirm what Farina et al (2004) suggested, namely that increasing the number of channels (and IED to 10 mm) would have resulted in higher ICC.

Initial value of FD has been found as the most repeatable parameter, with ICC values ranging from 0.91 (20% MVC) to 0.95 (60% MVC): this result can be partly explained by the very strong inter-subject variability (figures 4(a) and (b)). Nevertheless, the trial-to-trial variability expressed by the SEM was very small (9  ×  10⁻⁴), as also the MDC₉₅ (range 0.002–0.003) for both contraction levels (table 2). These findings suggest that initial value of FD is a reproducible variable which may be suitable in a clinical setting to estimate initial levels of MU synchronization in the biceps brachii.

Despite the very high relative test-retest reliability found for FD initial values during the 20% MVC, the rate of change of FD showed only a moderate inter-subject variability (ICC  =  0.67). This behavior may be explained by the fact that during a low level isometric contraction at 20% MVC the level of synchronization remains almost constant, just as with the muscle fiber CV (Beretta-Piccoli et al 2015). Two other studies conducted at an intensity of 20% (Contessa et al 2009) and 10% MVC (Semmler et al 2000) did not report an increase in MU synchronization during the time course of the contraction.

Moreover, just like CV slope at 20% MVC (figure 3(c)), the Bland & Altman plot of FD slope (figure 4(c)) resulted in a narrow value distribution, which suggests low inter-subject variability. Additionally, MDC₉₅, which estimates the smallest amount of change that can be detected by the measure, and SEM, which estimates the response stability, are both very low, respectively 0.001 and 0.002% s⁻¹, thus supporting the fact that the low test-retest reliability is probably not due to a measurement issue. Subsequently due to the fact that the 60% MVC contraction was fatiguing, as highlighted by a much more negative FD slope (which is an indicator of an increased MU synchronization (Beretta-Piccoli et al 2015, Boccia et al 2016a)), the relative test-retest reliability was higher (ICC  =  0.82) with respect to the 20% MVC contraction. Very low values of SEM (0.004) and MDC₉₅ (0.012% s⁻¹), supported by high inter-subject variability as showed in the Bland & Altman plot (figure 4(d)) suggest that FD rate of change during fatiguing contractions is a reliable measure of MU synchronization in biceps brachii.

Attention should be paid when assessing other muscles, as subject-dependent factors (e.g. fat layers and skin properties) may influence the FD of the signal but not the rate of change of FD (Mesin et al 2009). As a result, single values of the FD will not necessarily reflect the actual MU synchronization level; nevertheless, FD slope provides a reliable estimate of the actual change in MU synchronization level.