Spike shape analysis for the surface and needle electromyographic interference pattern

Biomedical Signal Processing and Control 36 (2017) 1–10

Contents lists available at ScienceDirect

Biomedical Signal Processing and Control journal homepage: www.elsevier.com/locate/bspc

Spike shape analysis for the surface and needle electromyographic interference pattern Lara A. Green a,∗ , Anita Christie b , David A. Gabriel a a b

Faculty of Applied Health Sciences, Brock University, St. Catharines, ON L2S 3A1 Canada Department of Human Physiology, University of Oregon, Eugene, OR 97403, United States

a r t i c l e

i n f o

Article history: Received 8 October 2016 Received in revised form 27 January 2017 Accepted 22 March 2017 Keywords: Biceps brachii Isometric force Motor unit Elbow flexion Muscle activity Electromyography

a b s t r a c t Introduction: The ability of surface and needle electromyographic (EMG) spike shape measures to match changes in motor unit recruitment, firing rate, and synchronization during force gradation, were compared. The purpose of the study was to determine the force level at which the surface EMG spike shape measures no longer parallel their indwelling analogues. Secondarily, the impact of the noise rejection criterion on the sensitivity of the spike shape measures was examined. Methods: Maximal isometric elbow flexion ramp contractions were performed while recording surface and needle EMG from the biceps brachii. Spike shape measures were calculated in 500 ms epochs over the duration of the ramp contraction. The spike threshold for needle EMG spike detection was varied to examine the effect of the algorithm’s selectivity. The pattern of change across force levels between surface and needle EMG measures was compared. Results: Spike detection resulted in the same pattern of change for both surface and needle amplitude measures over the gradation of force. Frequency measures and mean number of peaks per spike (MNPPS) were affected by electrode-source distance and spike threshold. Surface and needle frequency measures changed in parallel to 50% MVC while the MNPPS plateaued at 50–55% MVC. Discussion: Spike shape analysis of surface EMG can track changes in the interference pattern produced by recruitment and rate-coding up to 50% MVC. © 2017 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction Surface electromyography (sEMG) is a non-invasive way to assess global muscle activity, but is insufficient for recording motor unit activity patterns. Cancellation and alteration of the signal as it is recorded from the muscle through fascia, fat, and skin are some of the main factors that limit the physiological interpretation of the sEMG signal [1,2]. Indwelling EMG, recorded through wires or needles inserted directly into the muscle, allows for the identification of individual motor units. However, indwelling recordings are more invasive, technically demanding, and have a limited pick-up

Abbreviations: ANOVA, analysis of variance; EMG, electromyography; MDF, median power frequency; MNPPS, mean number of peaks per spike; MPF, mean power frequency; MSA, mean spike amplitude; MSD, mean spike duration; MSF, mean spike frequency; MSS, mean spike slope; MVC, maximal voluntary contraction; RMS, root mean square; sEMG, surface electromyography. ∗ Corresponding author at: Faculty of Applied Health Sciences, Brock University, 1812 Sir Isaac Brock Way, St. Catharines, ON L2S 3A1, Canada. E-mail addresses: [email protected] (L.A. Green), [email protected] (A. Christie), [email protected] (D.A. Gabriel).

volume [3,4]. As a result, the needle must be re-inserted in different locations of the muscle to obtain a sufficient number of motor units to collect a representative sample [5,6]. Spike shape analysis of the sEMG signal has the advantage of identifying motor unit activity patterns from the non-invasive sEMG signal which comprises a larger pick-up volume compared to the analysis of indwelling recordings [7]. The analysis technique detects changes in recruitment, firing rate, and synchronization through pattern classification of alterations in five discrete measures extracted from the sEMG interference pattern [8]. The five measures are mean spike amplitude (MSA), mean spike slope (MSS), mean spike frequency (MSF), mean spike duration (MSD), and the mean number of peaks per spike (MNPPS). Experimental and modelling studies have shown that these five measures change in a systematic way that is unique for each of the three different motor unit activity patterns (recruitment, firing rate, and synchronization) [9–11]. It is important to emphasise that spike shape analysis does not identify individual motor units, or calculate their discharge rate and recruitment. Rather, changes in the shape of individual spikes of the interference pattern have been used to successfully identify changes in motor unit activity patterns

http://dx.doi.org/10.1016/j.bspc.2017.03.006 1746-8094/© 2017 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/ 4.0/).

2

L.A. Green et al. / Biomedical Signal Processing and Control 36 (2017) 1–10

with the gradation in force [9,12] and have previously been used to discriminate between healthy controls versus a patient population [10,13]. In this respect, spike shape analysis is consistent with the call for a non-invasive approach to the detection of neuromuscular disorders [14], however, further experimental validation is necessary to define the limitations of surface EMG spike shape measures. Our previous modelling and simulation work has demonstrated that the ability of spike shape measures to detect changes in the sEMG interference pattern (sensitivity) during the gradation of force, is affected by both inter-electrode distance and electrode-source distance [11]. With larger inter-electrode and electrode-source distances, the spike shape measures plateau earlier in the force gradation process (i.e., at lower percentages of maximal voluntary contraction, % MVC). Experimentally, electrodesource distance effects may be evaluated by comparing indwelling and surface EMG recordings, which also incorporates differences in inter-electrode distance. Since indwelling needle electrodes have the smallest electrode-source and inter-electrode distances and presumably the greatest sensitivity to changes in muscle activity [15], the primary purpose of this paper was to experimentally determine the force level (%MVC) at which the surface EMG spike shape measures no longer parallel their indwelling analogues (i.e., become insensitive). A secondary purpose was to determine the impact of the noise rejection criterion on the sensitivity of the spike shape measures to changes in indwelling EMG interference patterns during the force gradation process. To date, the detection of individual spikes in the surface EMG interference pattern is based on both upward and downward deflections of a spike exceeding the 95% confidence interval for baseline noise [9]. The threshold level used to determine the number of turns and turns amplitude for the indwelling EMG signal has been shown to impact the ability to detect differences between groups in underlying motor unit activity patterns [16], and the EMG-to-force relationship during step isometric contractions from 0 to 100% MVC [17]. It is important therefore to determine the impact of the noise rejection criterion for indwelling EMG spike shape measures. To this end, spike shape measures from biceps brachii needle and surface EMG interference patterns were calculated over a 0–100% MVC isometric ramp contraction. The biceps brachii was selected for two reasons. First, motor unit recruitment and ratecoding during the gradation of force are well-known for this muscle [12,18,19]. Second, there are only two studies to date that have quantified the same measures from simultaneous recordings of surface and indwelling EMG: Preece et al. [20] compared surface and indwelling EMG of the tibialis anterior, but only in 100% MVC contractions, while Philipson and Larsson [17] studied step isometric contractions of the biceps brachii in 20% MVC increments up to 100% MVC. Thus, only other study available for a direct comparison is based on the biceps brachii. 2. Materials and methods Eleven participants (8 males, 3 females) between the ages of 18 and 45 participated in the present study. Participants reviewed and signed the informed consent document, which detailed all procedures and included ethical approval from the Human Ethics Review Board at the University of Massachusetts, Amherst. Testing for each participant was completed within a single session. 2.1. Experimental set up Participants were seated in a chair for testing and rested their right elbow on a platform directly in front of them so that the shoul-

der and elbow were flexed to approximately 90◦ . The wrist was in a neutral position with a custom-moulded fiberglass splint on the anterior side of the right wrist and forearm. Isometric elbow flexion was performed by pulling against the strain gauge force transducer (Interface Model MB-250, Scottsdale, AZ) attached to the fiberglass splint. The force signal was amplified and low-pass filtered at 10 Hz (DataQ PM-100, DataQ Instruments, Akron, OH), then sampled at 50 Hz using a 16-bit A/D converter (DT-322, Data Translation, Marlboro, MA). The force signals were also sent to a second computer to be sampled at 25,600 Hz using another 16-bit A/D converter (NIDAQ PCI-6251, National Instruments, Austin, TX) and stored offline for analysis. 2.2. Electromyography Electromyography was recorded from the short head of the biceps brachii. Surface EMG was recorded in a bipolar electrode configuration. Ag/AgCl electrodes (3-mm diameter) were placed 2cm apart on the belly of the biceps. The sEMG signals were amplified and band-pass filtered at 10–2000 Hz using a Dantec Counterpoint Electromyograph (Dantec Electronik Medicinsk, Sklovlunde, Denmark). The sEMG signals were initially sampled at 25,600 Hz and later downsampled to 2560 Hz for analysis using MATLAB (Mathworks Inc., Natick, MA). Needle EMG was recorded using a quadrifilar needle consisting of a 25-gauge stainless steel cannula housing 4 platinum-iridium wires 50 ␮m in diameter. This provided three channels of differential recordings. The needle EMG was amplified as required and bandpass filtered at 1–10 kHz using a Dantec Counterpoint Electromyograph (Dantec Electronik Medicinsk, Sklovlunde, Denmark). The needle signals were initially sampled at 25,600 Hz and later upsampled to 51,200 Hz to maximize resolution for motor unit identification in previous analyses [12] using MATLAB (Mathworks Inc., Natick, MA, USA). 2.3. Protocol The testing session began with three “hard and fast” maximal voluntary contractions lasting 5-s each to determine each participant’s maximal force. There were 2 min of rest between each contraction. Using the highest value of the three contractions a force trajectory was presented to the participants as a ramp to maximal force at a rate of 10% MVC/s. Participants completed 10 ramp contractions to maximum force with three minutes rest between each contraction. 2.4. Data reduction A total of 10 trials were selected to be used for analysis. One ramp contraction was selected per participant with the exception of one participant for which no contraction was usable. Selection criteria for the contractions included: reaching 100% MVC force level, minimal EMG baselines, and the presence of an needle EMG interference pattern. Once the 10 trials had been selected, one needle EMG channel was selected from the three channels based on interference pattern quality. While the channel was deemed excellent for interference pattern analysis of the particular trial, it had been unusable for motor unit identification in a previous study [12]. The onset of each contraction was the point at which the sEMG activity exceeded a noise threshold, which was set at the baseline sEMG activity plus 1.96 times the standard deviation. Each trial was then visually inspected for accuracy of sEMG onset. Surface and needle EMG data were then assessed using 22 epochs, 500 ms in duration, starting from the onset of the contraction. For each 500 ms epoch traditional and spike shape measures were calculated for both surface and needle EMG. Traditional measures included

L.A. Green et al. / Biomedical Signal Processing and Control 36 (2017) 1–10

3

Fig 1. Representative figure from one participant including: force (top black trace), needle electromyography (bottom black trace), and surface electromyography (blue trace). The surface EMG was further zoomed in to display the motor unit action potentials identifiable at low levels of force (approximately 0–15% MVC). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

root-mean-square (RMS) amplitude and mean power frequency (MPF). Spike shape measures were calculated from a custom algorithm, which identified EMG spikes exceeding a predetermined spike threshold criterion. The surface EMG spike threshold was the same as that for onset detection (baseline noise ±1.96* standard deviation). This spike threshold criterion was maintained for the sEMG spike shape measures, as previous work has demonstrated that sensitivity to changes in force is best achieved with the 95% confidence interval [8]. The interaction between electrode-source distance and spike threshold criterion was assessed by adjusting the needle EMG spike threshold criterion. Since tissue filtering associated with the sEMG signal is fixed, the needle EMG spike threshold criterion was manipulated to determine the level at which the spike shape measures for the two recordings were most comparable. Therefore, the needle EMG spike threshold was calculated as the baseline mean amplitude plus the standard deviation multiplied by a varying multiplication factor starting at 0 and going to 2.5, in increments of 0.25. Preliminary analysis revealed that the spike threshold for which the surface and needle measures were most comparable was the 95% confidence interval for surface measures and the 99% confidence interval for needle measures. The spike shape algorithm, described in detail elsewhere [8] identifies a spike as an upward and downward deflection crossing zero and exceeding the spike threshold criterion. From each spike we can calculate the peak-to-peak amplitude, slope, duration, and frequency of occurrence. Furthermore, a pair of upward and downward deflections within a spike, that does not constitute a separate spike (as defined by the spike threshold criterion), is referred to as a peak. All data analysis was performed offline using MATLAB (Mathworks Inc., Natick, MA, USA). 2.5. Statistical analysis Repeated measures analysis of variance (ANOVA) were used to assess the difference between surface and needle EMG across

epochs, for all measures. Orthogonal polynomials were also used for trend comparisons, which assessed the similarity between the surface and needle EMG spike shape measures across epochs. A comparison of the surface versus needle EMG trends across epochs assessed the ability of the spike shape measure to accurately reflect the physiological changes occurring with the gradation of force (i.e., sensitivity to physiological changes). We have successfully used this statistical approach to compare changes in spike shape measures obtained with monopolar versus bipolar surface EMG recordings during step isometric contractions from 40 to 100% MVC [21]. A trend component was considered non-trivial only if it accounted for more than 15% of the total variance as is consistent with a medium effect-size [22]. The alpha was set at the 0.05 ® probability level. All statistical procedures were performed in SAS (SAS Institute Inc.; Cary, NC, USA).

3. Results A representative trial, including force, surface EMG, and a single channel of needle EMG, is displayed in Fig. 1. A depiction of the individual motor unit action potentials that can be seen from the surface EMG signal at low levels of force (approximately 0–15% MVC), are displayed from a representative trial in the bottom trace of Fig. 1. As expected, there was a significant difference in the overall magnitude of surface versus needle EMG (note the scales in Fig. 1). Representative trials demonstrating the first and tenth ramp contraction from one participant is displayed in Fig. 2, demonstrating that fatigue was not a factor in the selection of the ten trials. The focus of this paper is on the pattern changes in the specific measures across force levels as detailed below. Similarly, the traditional (RMS and MPF) and spike shape measures are presented together for visual comparison only as they were not statistically compared. In Figs. 3–7 , the darkest line for spike shape measures of needle EMG (right panels) corresponds to the 99% confidence interval for noise. The blue line represents the lowest threshold, corresponding

4

L.A. Green et al. / Biomedical Signal Processing and Control 36 (2017) 1–10

Fig. 2. Representative figure from one participant displaying the first (black) and tenth (grey) ramp contraction including force (top), needle EMG (middle), and surface EMG (bottom).

Fig. 3. Surface (left panel) and needle (right panel) electromyography amplitude measures. The red lines are the mean and standard error of the root-mean-square (RMS) amplitude. The thick black lines are the mean spike amplitude (MSA) calculated with a 95% and 99% confidence interval baseline spike threshold for surface and needle, respectively. For needle MSA, the blue line represents the lowest spike threshold criteria, which is the mean of baseline activity. The grey lines are in increments of 0.25 times the standard deviation plus the mean of baseline activity (between 0 and 2.57). The vertical grey window is presented as a common reference point and corresponds to the plateau of needle MSA and MSF between approximately 70–85% MVC.

to the mean of baseline activity (noise). The impact of increasing threshold levels on the spike shape analysis for needle EMG is visually depicted as additional grey lines in increments of 0.25 times

the standard deviation added to the mean baseline. The vertical grey window corresponds to the plateau of needle MSA and MSF between approximately 70–85% MVC to facilitate the comparison

L.A. Green et al. / Biomedical Signal Processing and Control 36 (2017) 1–10

5

Fig. 4. Surface (left panel) and needle (right panel) electromyography mean spike slope (MSS). The thick black lines are the mean spike slope (MSS) calculated with a 95% and 99% confidence interval baseline spike threshold for surface and needle, respectively. The blue line (lowest spike threshold criteria), grey lines (incremental spike threshold criteria), and grey vertical window, correspond to the descriptions given in text and in Fig. 3. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 5. Surface (left panel) and needle (right panel) electromyography frequency measures. The red lines are the mean and standard error of the mean power frequency (MPF) amplitude. The thick black lines are the mean spike frequency (MSF) calculated with a 95% and 99% confidence interval baseline spike threshold for surface and needle, respectively. The blue line (lowest spike threshold criteria), grey lines (incremental spike threshold criteria), and grey vertical window, correspond to the descriptions given in text and in Fig. 3.

6

L.A. Green et al. / Biomedical Signal Processing and Control 36 (2017) 1–10

Fig. 6. Surface (left panel) and needle (right panel) electromyography mean spike duration (MSD). The thick black lines are the mean spike duration (MSD) calculated with a 95% and 99% confidence interval baseline spike threshold for surface and needle, respectively. The blue line (lowest spike threshold criteria), grey lines (incremental spike threshold criteria), and grey vertical window, correspond to the descriptions given in text and in Fig. 3. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 7. Surface (left panel) and needle (right panel) electromyography mean and standard error of the mean number of peaks per spike (MNPPS). The thick black lines are the mean number of peaks per spike (MNPPS) calculated with a 95% and 99% confidence interval baseline spike threshold for surface and needle, respectively. The blue line (lowest spike threshold criteria), grey lines (incremental spike threshold criteria), and grey vertical window, correspond to the descriptions given in text and in Fig. 3. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

L.A. Green et al. / Biomedical Signal Processing and Control 36 (2017) 1–10

of changes in needle and surface EMG. Specifically, to compare the point at which surface versus needle measures plateau during the force gradation process.

3.1. Amplitude The ramp increase in force, plateau at maximum, and abrupt termination results in a similar pattern of change for both the surface and needle EMG amplitude measures (see Fig. 3). The RMS amplitude showed a significant interaction between electrode type across epochs (F(21,378) = 31.32, p < 0.0001). Table 1 details the ANOVA interaction and the variance accounted for by the first four trend components. The RMS amplitude measures for both electrode type exhibited significant trend components up to the quadratic degree. However, percent variance accounted for by the linear and quadratic trend components differed between for surface versus indwelling recordings. For example, the linear trend component was strongest in the surface RMS amplitude (93.5%) while the needle had a greater contribution from the quadratic component (18.3%). The cubic trend was similar for both electrode types at approximately 5%, but not deemed to be a major component of the overall pattern. The MSA showed a significant interaction between electrode type across epochs (F(21,378) = 24.24, p < 0.0001). Orthogonal polynomial testing further revealed significant differences between electrode types (Fig. 3). The linear component accounted for the greatest variance across epochs in both electrode types, but was higher in surface EMG (93.6%) compared to needle EMG (73.2%). The quadratic component was minimal in sEMG (0.5%) but significant in needle EMG (21.5%). The cubic component was similar for both electrode types at approximately 5%. Increasing the spike threshold did not alter the pattern of change for needle MSA, only the overall magnitude. The MSS (Fig. 4) measure followed the same pattern of change as MSA, so those results will not be presented for the sake of brevity.

3.2. Frequency The MPF showed a significant interaction between electrode type across epochs (F(21,378) = 3.00, p < 0.0001). Table 2 details the ANOVA interaction and the variance accounted for by the first four trend components. Orthogonal polynomial testing revealed that surface EMG MPF was dominated by a quadratic trend that accounted for 85.0% of the variance (Fig. 5). The means begin and end near the same point while peaking between 55 to 75% of MVC. Overall, needle MPF was dominated by a linear trend that accounted for 74.8% of the variance. The plateau in needle MPF, starting at approximately 50% MVC, contributed a quadratic trend component accounting for 15.5% of the variance. The MSF showed a significant interaction between electrode type across epochs (F(21,378) = 14.14, p < 0.0001). Both electrode types exhibited linear and quadratic trend components but the trend component that dominated the distribution of the variance was different for surface versus needle recordings (Fig. 5). While surface EMG exhibited a linear increase, it accounted for less than half (27.0%) of the variance. For needle EMG the linear component accounted for the greatest variance (64.4%). The quadratic component accounted for a majority of the variance in the surface MSF (67.7%) but only 34.6% for needle MSF. The MSD (Fig. 6) is used in the calculation of MSF and therefore followed the exact inverse pattern of change as MSF. The results for MSD are therefore omitted for the sake of brevity.

7

3.3. MNPPS The MNPPS did not show a significant interaction between electrode type across epochs (F(21,378) = 0.94, p = 0.534). Table 2 details the ANOVA interaction and the variance accounted for by the first four trend components. The two electrode types were not statistically different on any of the orthogonal trends. Both needle and surface recordings had significant quadratic trend components, accounting for 87.8% of the variance for the needle and 36.3% for surface (Fig. 7). The quadratic pattern for surface MNPPS was part of an overall linear decrease accounting for 47.4% of the variance. In contrast, the change in needle MNPPS was almost entirely quadratic, with the first and last means beginning and ending at the same point. The minimum needle MNPPS occurred around 55% MVC. 4. Discussion The present study compared spike shape measures of surface and needle EMG interference patterns. In general, the surface and needle spike shape measures changed in parallel across force levels up to approximately 50% MVC. The changes in EMG measures followed known motor unit activity patterns associated with the gradation of force. However, the pattern of change was affected by electrode-source distance, inter-electrode distance, and spike threshold criteria used for spike detection. In the following paragraphs, we will explain how physical and technical aspects of EMG recording interact with the underlying physiology to produce the observed indwelling and surface interference patterns, as revealed by both experimental techniques and modelling and simulations studies. 4.1. Amplitude measures The needle and surface RMS and MSA measures demonstrated nearly identical patterns of change across force levels (Fig. 3). The measures exhibited a linear increase until approximately 75–80% MVC, followed by a slight plateau and subsequent decrease towards the end of the contraction. Philipson and Larsson [17] completed a study to similar the present work, using step isometric contractions of the elbow flexors, in 20% MVC increments from 0 to 100% MVC. The authors did not quantify the EMG-to-force relationship for surface and indwelling EMG measures, however, inspection of their figures reveals nearly identical results to ours. The overall curvilinear pattern for surface and indwelling EMG amplitude has been previously reported for ramp contractions [11,23,24]. The pattern of increase in EMG amplitude is consistent with a rapid increase in motor unit size, which lasts until approximately 80% MVC [23,25], coinciding with the motor unit recruitment range of the biceps brachii [18]. The decrease in the amplitude measures seen after approximately 85% of maximum force (see Fig. 3) can be explained by motor unit dropout due to the length of the contraction (10–11 s). Previous research has shown that high threshold motor units have a tendency to be de-recruited (i.e., to dropout) at high force levels [31]. The slow nature of the ramp at only 10% MVC per second and the plateau at maximal force may have provided the length of time necessary for fast-fatiguing motor units to dropout. A decline in force does not accompany the dropout in motor units, which can be explained by the increase MSF, MPF, and MNPPS, as firing rates increase to compensate for the motor unit dropout. The concomitant decrease in MSS at high force levels coincides with the increase in MSD, which is sufficient enough to affect the slope of the spikes. Since the high selectivity of the indwelling electrode records from a relative few motor units, compared to the surface electrode, indwelling MSA is more affected by motor unit dropout compared

8

L.A. Green et al. / Biomedical Signal Processing and Control 36 (2017) 1–10

Table 1 The statistics are presented for root-mean-square amplitude, mean spike amplitude, and mean spike slope. The ANOVA F-value for electrode (surface and indwelling) by epoch (22 epochs of 500 ms each) interaction is presented in the top row. Trend components from the orthogonal polynomials analysis is presented as the percent variance accounted for by the linear, quadratic, cubic, and quartic components. *p < 0.05, ** p < 0.001.

Electrode x Epoch Interaction

% Variance Accounted 1 Linear 2 Quadratic 3 Cubic 4 Quartic

RMS

MSA

MSS

31.32**

24.24**

5.02**

Surface

Indwelling

Surface

Indwelling

Surface

Indwelling

93.5%** 0.9% 5.2%* 0.3%

76.1%** 18.3%* 5.1%* 0.02%

93.6%** 0.5% 5.2%* 0.4%

73.2%** 21.5%* 4.7%* 0.03%

90.2%** 2.8%* 6.6%* 0.07%

48.6%* 46.1%* 2.5% 0.3%

Table 2 The statistics are presented for mean power frequency, mean spike frequency, mean spike duration, and mean number of peaks per spike. The ANOVA F-value for electrode (surface and indwelling) by epoch (22 epochs of 500 ms each) interaction is presented in the top row. Trend components from the orthogonal polynomials analysis is presented as the percent variance accounted for by the linear, quadratic, cubic, and quartic components. *p < 0.05, ** p < 0.001. MPF Electrode x Epoch Interaction

% Variance Accounted 1 Linear 2 Quadratic 3 Cubic 4 Quartic

MSF

3.00**

MSD

14.14**

MNPPS

6.28**

0.94

Surface

Indwelling

Surface

Indwelling

Surface

Indwelling

Surface

Indwelling

3.8% 85.0%* 3.7%* 0.3%

3.8% 85.0%* 3.7%* 0.3%

27.0%* 67.7%** 3.7%* 0.001%

64.4%* 34.6%** 0.4% 0.04%

0.5% 91.8%* 5.2%* 0.01%

97.5%* 0.07% 1.0% 0.003%

47.4%* 36.3%* 5.1%* 3.6%*

1.2% 87.8%* 0.6% 0.6%

to its surface counterpart. Furthermore, the greater MSD associated with surface recordings allows for temporal summation to maintain the surface MSA. It should be noted that the steep decrease between the last 2 points (10.5–11 s) coincides with the release of the ramp contraction. The similarity of patterns between surface and needle MSA is explicable as cancellation and electrode-source distance have been shown to affect the magnitude, but not the pattern of change with increasing force [11,26]. Spike threshold level for needle EMG affected only the magnitude, but not pattern, of the amplitude measures. Fig. 3 (right panel) demonstrates the ranging threshold levels calculated from the baseline mean activity plus the standard deviation multiplied by a multiplication factor of 0 (mean baseline activity alone; blue line) to 2.57 (mean baseline activity plus 2.57 times the standard deviation; black line). A decrease in the multiplication factor resulted in a lower mean amplitude, as smaller spikes were included in the calculation [16]. This however, did not affect the overall shape of either amplitude measure (MSA and MSS). 4.2. Frequency measures There was an increase in surface MPF and MSF until approximately 60% MVC, then a gradual decrease occurred. Needle MSF followed a similar pattern but continued to increase until approximately 70% MVC, before decreasing (see vertical grey shading in Fig. 5). Philipson and Larssson [17] studied the number of zero crossings, which is correlated with the frequency content of the signal [27]. Inspection of their zero-crossings figures reveals nearly identical results as exhibited in our surface and indwelling frequency data, respectively [17]. The MPF was higher at low-force levels compared to MSF, in both surface and indwelling recordings, due to the presence of small, high frequency motor unit action potentials and/or high frequency noise components [28] as seen in Fig. 1 (bottom trace). In the absence of fatigue, frequency measures parallel increases in muscle fiber conduction velocity associated with the recruitment of higher threshold motor units [23,24,29,30]. Rate-coding is continuous with the gradation of muscle force, but motor unit recruitment is the dominant factor in muscles with a broad recruit-

ment range, such as the biceps brachii [19,31]. Compared to muscles with a narrow recruitment range (50% MVC) such as the first dorsal interosseous, the biceps brachii has lower firing rates within its broader recruitment range [19]. Lower firing rates and shorter EMG wave durations associated with highly selective indwelling recordings [15,32] reduce the probability of temporal overlap [33]. As a result, the impact of rate-coding on the temporal summation of waveforms in the indwelling needle EMG interference was secondary. A different indwelling EMG frequency-force relationship can be expected for smaller muscles with a narrow recruitment range (50% MVC), such as first dorsal interosseous [19,31]. Smaller muscles depend on rate-coding earlier during the gradation of force and have higher firing rates than larger muscles [19,31]. Since higher firing rates can increase the probability of temporal overlap [34], the impact of summation on the frequency content of the indwelling EMG signal may be more pronounced below 80% MVC. Thus, the present results do not generalize to small muscles with a narrow recruitment range and they need to be investigated in future work. The pattern of change for MPF is nearly identical to needle MSF at the lowest spike threshold level, which provides unique insight into the interaction between electrode-source distance and spike threshold criteria in the following way. The lowest threshold criteria for needle MSF included the smallest detectable spikes, similar to needle MPF, which includes the entire signal. As a result, the temporal overlap between motor unit action potentials associated with both recruitment and rate-coding begins immediately. There was a progressive decrease in needle MPF, which plateaued towards the end of the recruitment range. At this point rate-coding continued to produce temporal overlap but not to the same degree. In support of this idea, several modelling studies have documented the reduction in MPF associated with increased rate-coding at the highest force levels [34,35]. The spike threshold criteria for both surface and needle MSF allows for the detection of discrete spikes without overlap. In the case of MSF, the recruitment of additional motor units is observed as more spikes per unit time until temporal overlap becomes significant, which occurs earlier for surface (∼60% MVC) versus needle (∼70% MVC) recordings. Temporal overlap becomes significant for

L.A. Green et al. / Biomedical Signal Processing and Control 36 (2017) 1–10

surface EMG at a lower percentage of MVC, compared to needle EMG, due to greater electrode-source and inter-electrode distances. Electrode-source distance dominates and dramatically decreases surface EMG amplitude, while both electrode-source and interelectrode distances increase duration [11,15,36,37]. The increase in surface EMG wave duration increases the probability of temporal overlap [33]. Furthermore, the effect of temporal overlap is more pronounced in the surface MPF measure due to inclusion of the entire signal, which also results in decreased sensitivity to changes in the interference pattern compared to surface MSF.

4.3. Mean number of peaks per spike A peak is a pair of upward and downward deflections within a spike that does not constitute a separate spike. Magora and Gonen [38] detailed the importance of peaks as being the only motor unit characteristic that is preserved in an interference pattern. Each individual firing of a motor unit will be evident as either a spike or peak, depending on amplitude, barring the rare occurrence of two motor units with similar parameters firing at exactly the same time [38]. Magora and Gonen [38,39] showed that synchronized motor unit firing patterns decrease needle MNPPS, and that asynchronous firing patterns associated with higher firing rates increase needle MNPPS. In the present study we observed a pronounced increase in the needle MNPPS towards the end of the recruitment range. These experimental results refute earlier modelling work which suggested that synchronization dominates the frequency content of the surface EMG signal at high force levels [35]. Taken together with the results for needle MNPPS, the decrease in needle MSF after the plateau is due to rate-coding. The surface MSF exhibited the same effects, but electrode-source distance increased the duration of surface action potentials and probability of temporal overlap [15,34,36]. As a result, the reduction in surface MSF began earlier than needle. There was only a mild increase in surface MNPPS (9%) compared to needle MNPPS (26%). The difference between the two measures is easily accounted for by electrode-source distance [11]. Surface MNPPS started to plateau at ∼50% MVC. While changes in surface MNPPS beyond this point mirrored the needle measure, it was very mild. Interpretation of changes in the MNPPS within the context of the frequency measures, indicates the sensitivity of surface EMG spike shape analysis decreases dramatically after 50% MVC. Previous work has demonstrated that spike shape analysis of the surface EMG signal could separate patients with repetitive strain injury exhibiting myopathic changes from those “at risk” for myopathic changes versus normal controls [10]. Step-isometric contractions were performed up to 70% MVC and spike shape analysis identified distinctly different changes in the surface interference pattern that was corroborated by quantification of needle motor unit activity [40]. The present results provide experimental data that partially supports modelling and simulation studies [41,42], outlining the limitations of the use of surface EMG to study neural strategies. Our results confirm that, above 50% MVC, the surface EMG interference pattern is insensitive to changes in motor unit activity patterns. However, spike shape measures of the surface EMG interference pattern exhibited systematic changes up to 50% MVC, consistent with known motor unit activity patterns during force gradation. We believe that the discrepancy between our experimental findings and modelling and simulation results is due the use of MPF, median power frequency (MDF), RMS amplitude, and mean amplitude value to examine the relationship between surface EMG and underlying neurophysiology. Our previous work has demonstrated quite clearly that these measures are highly susceptible to temporal summation, which decreases the ability to detect changes in the

9

interference pattern; a point that reinforced by the excellent work on cancellation [26]. Since MPF and RMS amplitude include all waveforms within the data analysis window, the sensitivity of these measures are decreased by greater susceptibility to temporal overlap, i.e., overlapping waveforms below the 95% confidence interval for baseline noise contribute to the pattern of change across force levels [43]. We define noise as low amplitude baseline signal from multiple sources: instrumentation, electromagnetic, cross-talk from other muscles, and/or contributions from distant motor unit action potentials [32,44]. The use of spike threshold criteria is especially critical considering that individual motor unit action potentials can be detected from the skin’s surface, even with standard surface electrodes, early in the force gradation process (ie., 30% MVC) [45]. Spike shape analysis can detect these low threshold, smaller motor unit action potentials against the background noise, and extract them for analysis. In contrast, the impact of low threshold, smaller motor unit action potentials is diluted by background noise when analysed by traditional time and frequency analyses. In support, Marusiak et al. [13] demonstrated that spike shape analysis could detect underlying changes in muscle control associated with Parkinson’s tremor while RMS amplitude and MPF failed to detect the differences between tremor and non-tremor EMG signals. 5. Conclusion Spike shape analysis of surface EMG is less sensitive to changes in underlying motor unit activity above 50% MVC, as surface MSF and MNPPS no longer parallel needle activity beyond this point. Differences in the pattern of change across force levels between surface and needle spike shape measures may be explained by the interaction of electrode-source and inter-electrode distances and spike threshold criteria. Spike threshold criteria altered the absolute magnitude of MSA but not its pattern of change across force levels. In contrast, spike threshold altered both the magnitude and pattern of change in MSF and MNPPS. Acknowledgments This work was funded by an operating grant to David A. Gabriel from the Natural Sciences and Engineering Research Council of Canada#227790. None of the authors have potential conflicts of interest to be disclosed. References [1] K.G. Keenan, D. Farina, R. Merletti, R.M. Enoka, Amplitude cancellation reduces the size of motor unit potentials averaged from the surface EMG, J. Appl. Physiol. 100 (2006) 1928–1937, http://dx.doi.org/10.1152/japplphysiol. 01282. [2] D. Farina, R. Merletti, R.M. Enoka, The extraction of neural strategies from the surface EMG: an update, J. Appl. Physiol. 117 (11) (2014) 1215–1230, http:// dx.doi.org/10.1152/japplphysiol.00162.2014. [3] J.R. Daube, D.I. Rubin, Needle electromyography, Muscle Nerve 39 (2009) 244–270, http://dx.doi.org/10.1002/mus.21180. [4] A.J. Boon, J.T. Gertken, J.C. Watson, R.S. Laughlin, J.A. Strommen, L. Michelle, Hematoma risk after needle EMG, Muscle Nerve 44 (2011) 439–440, http:// dx.doi.org/10.1002/mus.22227. [5] J.A. Strommen, J.R. Daube, Determinants of pain in needle electromyography, Neurophysiol. Clin. 112 (2001) 1414–1418, http://dx.doi.org/10.1016/s13882457(01)00552-1. ´ Size of motor unit potential sample, Muscle Nerve 27 [6] S. Podnar, M. Mrkaic, (2003) 196–201, http://dx.doi.org/10.1002/mus.10310. [7] D. Farina, F. Negro, M. Gazzoni, R.M. Enoka, Detecting the unique representation of motor-unit action potentials in the surface electromyogram, J. Neurophysiol. 100 (3) (2008) 1223–1233, http://dx.doi. org/10.1152/jn.90219.2008. [8] D.A. Gabriel, Reliability of SEMG spike parameters during concentric contractions, Electromyogr. Clin. Neurophysiol. 40 (2000) 423–430. [9] D.A. Gabriel, S.M. Lester, S.A. Lenhardt, E.D.J. Cambridge, Analysis of surface EMG spike shape across different levels of isometric force, J. Neurosci.

10

[10]

[11]

[12]

[13]

[14]

[15]

[16] [17]

[18]

[19]

[20]

[21] [22] [23]

[24]

[25]

[26]

L.A. Green et al. / Biomedical Signal Processing and Control 36 (2017) 1–10 Methods 159 (2007) 146–152, http://dx.doi.org/10.1016/j.jneumeth.2006.07. 004. K.M. Calder, D.A. Gabriel, L. McLean, Differences in EMG spike shape between individuals with and without non-specific arm pain, J. Neurosci. Methods 178 (2009) 148–156, http://dx.doi.org/10.1016/j.jneumeth.2008.11.015. D.A. Gabriel, A. Christie, J.G. Inglis, G. Kamen, Experimental and modeling investigation of surface EMG spike analysis, Med. Eng. Phys. 33 (2011) 427–437, http://dx.doi.org/10.1016/j.medengphy.2010.11.010. A. Christie, J.G. Inglis, G. Kamen, D.A. Gabriel, Relationships between surface EMG variables and motor unit firing rates, Eur. J. Appl. Physiol. 107 (2009) 177–185, http://dx.doi.org/10.1007/s00421-009-1113-7. ´ J. Marusiak, R. Andrzejewska, D. Swiercz, K. Kisiel-sajewicz, A. Jaskólska, A. Jaskólski, Spike shape analysis of electromyography for parkinsonian tremor evaluation, Muscle Nerve 52 (6) (2015) 1096–1098. M.J. Zwarts, G. Drost, D.F. Stegeman, Recent progress in the diagnostic use of surface EMG for neurological diseases, J. Electromyogr. Kinesiol. 10 (2000) 287–291, http://dx.doi.org/10.1016/s1050-6411(00)00020-1. D.W. Stashuk, Decomposition and quantitative analysis of clinical electromyographic signals, Med. Eng. Phys. 21 (6) (1999) 389–404, http://dx. doi.org/10.1016/s1350-4533(99)00064-8. K. Hirose, M. Uono, Noise in quantitative electromyography, Electromyogr. Clin. Neurophysiol. 25 (1985) 341–352. L. Philipson, P.G. Larsson, The electromyographic signal as a measure of muscular force: a comparison of detection and quantification techniques, Electromyogr. Clin. Neurophysiol. 28 (1988) 141–150. C.G. Kukulka, H.P. Clamann, Comparison of the recruitment and discharge properties of motor units in human brachial biceps and adductor pollicis during isometric contractions, Brain Res. 219 (1) (1981) 45–55, http://dx.doi. org/10.1016/0006-8993(81)90266-3. K. Seki, M. Narusawa, Firing rate modulation of human motor units in different muscles during isometric contraction with various forces, Brain Res. 719 (1) (1996) 1–7. A.W. Preece, H.S. Wimalaratna, J.L. Green, E. Churchill, H.M. Morgan, Non-invasive quantitative EMG, Electromyogr. Clin. Neurophysiol. 34 (2) (1994) 81–86. D.A. Gabriel, Effects of monopolar and bipolar electrode configurations on surface EMG spike analysis, Med. Eng. Phys. 33 (9) (2011) 1079–1085. J. Cohen, Statistical Power Analysis for the Behavioural Sciences, 2nd ed., Lawrence Erlbaum Associates, Hillsdale, NJ, 1988. J.-Y. Hogrel, Use of surface EMG for studying motor unit recruitment during isometric linear force ramp, J. Electromyogr. Kinesiol. 13 (2003) 417–423, http://dx.doi.org/10.1016/s1050-6411(03)00026-9. P. Sbricolli, I. Bazzucchi, A. Rosponi, M. Bernardi, G. De Vito, F. Felici, Amplitude and spectral characteristics of biceps brachii sEMG depend upon the speed of isometric force gradation, J. Electromyogr. Kinesiol. 13 (2003) 139–147, http://dx.doi.org/10.1016/s1050-6411(02)00098-6. S.G. Boe, D.W. Stashuk, T.J. Doherty, Motor unit number estimates and quantitative motor unit analysis in healthy subjects and patients with amyotrophic lateral sclerosis, Muscle Nerve 36 (2007) 62–70, http://dx.doi. org/10.1002/mus.20784. K.G. Keenan, D. Farina, K.S. Maluf, R. Merletti, R.M. Enoka, Influence of amplitude cancellation on the simulated surface electromyogram, J. Appl. Physiol. 98 (2005) 120–131, http://dx.doi.org/10.1152/japplphysiol.00894. 2004.

[27] G.F. Inbar, O. Paiss, J. Allin, H. Kranz, Monitoring surface EMG spectral changes by the zero crossing rate, Med. Biol. Eng. Comp. 24 (1) (1986) 10–18. [28] R. Andrzejewska, A. Jaskólski, A. Jaskólska, M. Gobbo, C. Orizio, Electromyogram features during linear torque decrement and their changes with fatigue, Eur. J. Appl. Physiol. 114 (10) (2014) 2105–2117. [29] F. Lange, T.W. Van Weerden, J.H. Van der Hoeven, A new surface electromyography analysis method to determine spread of muscle fiber conduction velocities, J. Appl. Physiol. 93 (2002) 759–764, http://dx.doi.org/ 10.1152/japplphysiol.00594.2001. [30] S. Andreassen, L. Arendt-Nielsen, Muscle fibre conduction velocity in motor units of the human anterior tibial muscle: a new size principle parameter, J. Physiol. 391 (1987) 561–571, http://dx.doi.org/10.1113/jphysiol.1987, sp016756. [31] C.J. De Luca, R.S. LeFever, M.P. McCue, A.P. Xenakis, Behaviour of human motor units in different muscles during linearly varying contractions, J. Physiol. 329 (1) (1982) 113–128. [32] D. Stashuk, EMG signal decomposition: how can it be accomplished and used? J. Electromyogr. Kinesiol. 11 (3) (2001) 151–173. [33] P. Zhou, W.Z. Rymer, Factors governing the form of the relation between muscle force and the EMG: a simulation study, J. Neurophysiol. 92 (5) (2004) 2878–2886. [34] A.J. Fuglevand, D.A. Winter, A.E. Patla, Models of recruitment and rate coding organization in motor-unit pools, J. Neurophysiol. 70 (6) (1933) 2470–2488. [35] D.A. Gabriel, G. Kamen, Experimental and modeling investigation of spectral compression of biceps brachii SEMG activity with increasing force levels, J. Electromyogr. Kinesiol. 19 (2009) 437–448, http://dx.doi.org/10.1016/j. jelekin.2007.10.009. [36] N.T. Artu˘g, I. Goker, B. Bolat, O. Osman, E.K. Orhan, M.B. Baslo, The effect of recording site on extracted features of motor unit action potential, Comput. Methods Programs Biomed. 129 (2016) 172–185, http://dx.doi.org/10.1016/j. cmpb.2016.01.003. [37] A.J. Fuglevand, D.A. Winter, A.E. Patla, D. Stashuk, Detection of motor unit action potentials with surface electrodes: influence of electrode size and spacing, Biol. Cybern. 67 (2) (1992) 143–153. [38] A. Magora, B. Gonen, Computer analysis of the shape of spike from electromyographic interference pattern, Electromyography 10 (3) (1970) 261–271. [39] A. Magora, B. Gonen, Clinical evaluation of the analysis of the shape of electromyographic shape, Electromyography 12 (3) (1972) 255–265. [40] K.M. Calder, M.J. Agnew, D.W. Stashuk, L. McLean, Reliability of quantitative EMG analysis of the extensor carpi radialis muscle, J. Neurosci. Methods 168 (2) (2008) 483–493, http://dx.doi.org/10.1016/j.jneumeth.2007.10.008. [41] D. Farina, M. Fosci, R. Merletti, Motor unit recruitment strategies investigated by surface EMG variables? J. Appl. Physiol. 92 (1) (2002) 235–247. [42] D. Farina, A. Holobar, R. Merletti, R.M. Enoka, Decoding the neural drive to muscles from the surface electromyogram, Neurophysiol. Clin. 121 (10) (2010) 1616–1623. [43] H. Christensen, A. Fuglsang-Frederiksen, Power spectrum and turns analysis of EMG at different voluntary efforts in normal subjects? Electroencephalogr. Clin. Neurophysiol. 64 (6) (1986) 528–535. [44] E.A. Clancy, E.L. Morin, R. Merletti, Sampling: noise-reduction and amplitude estimation issues in surface electromyography, J. Electromyogr. Kinesiol. 12 (1) (2002) 1–16. [45] K.C. McIntosh, D.A. Gabriel, Reliability of a simple method for determining muscle fiber conduction velocity? Muscle Nerve 45 (2) (2012) 257–265.