Abstract
This study investigated the relative proportion of motor-unit action potentials that are uniquely represented in the simulated and experimental surface electromyogram (EMG). Two hundred motor units were simulated in a cylindrical anatomical system. Action potentials for each motor unit were generated with a model and then compared with those of other motor units. Pairs of motor units were considered indistinguishable and the motor units not uniquely represented in the surface EMG, when the difference in the mean energy for the pair of potentials was <5%. The anatomical conditions and recording configurations had a substantial influence on the percentage of motor units that could be uniquely identified in the simulated EMG. For example, a single monopolar channel could discriminate only 3.4% of motor units in the simulated population, whereas a system with 81 Laplacian channels arranged in a grid could discriminate 83.8% of the motor units under the same conditions. The simulation results were confirmed with populations of motor units recorded experimentally from the abductor digiti minimi muscle of eight healthy men. Furthermore, the relative proportion of uniquely identified motor units in the simulated signal was only moderately related to motor-unit size and distance from the electrodes. These results indicate the upper limit for detection of individual motor units from the surface EMG and show that a few channels of surface EMG recordings are not sufficient to study single motor units. The noninvasive identification of motor units from the surface EMG requires the use of multiple channels of information.
INTRODUCTION
The identification of action potentials belonging to individual motor units provides information about the motor output from the spinal cord. The summed electrical activity of the active motor units can be recorded with electrodes either inserted into the muscle or placed on the skin over the muscle. Although intramuscular recordings of motor-unit activity have been used for about 80 yr (Adrian and Bronk 1929), attempts to identify the discharge of individual motor units in the surface electromyogram (EMG) have been a more recent development (De Luca et al. 2006; Gazzoni et al. 2004; Hogrel 2003; Holobar and Zazula 2004; Kleine et al. 2000, 2007, 2008; Zazula and Holobar 2005). The advantages of using surface EMG recordings are that the technique is noninvasive and it may be able to detect a greater number of motor units than is currently possible with intramuscular recordings (Holobar et al. 2006).
One of the challenges with the surface EMG approach is that the low-pass filtering effect of the tissues interposed between the fibers and the electrodes causes surface-detected potentials of different motor units to have a similar shape (Dimitrov and Dimitrova 1974; Merletti et al. 1999; Stegeman et al. 2000). However, the recording of the surface EMG with multiple electrodes, which are known as multichannel systems, may enhance the capacity to discriminate the action potentials of separate motor units (Blok et al. 2002; Zwarts et al. 2004). This is accomplished by multichannel systems providing many recordings of motor-unit activity along the length of a muscle (Masuda et al. 1985) or over its surface area (Blok et al. 2002), which facilitates discrimination of the action potentials belonging to different motor units.
The improved discrimination capacity of multichannel systems compared with the classic bipolar recording of the surface EMG has been shown for one subject using an EMG decomposition algorithm (Kleine et al. 2007). Under some conditions, however, it is possible to discriminate motor-unit action potentials with only a few surface EMG channels (De Luca et al. 2006). Despite demonstrations that specific algorithms can decompose representative recordings (De Luca et al. 2006; Gazzoni et al. 2004; Holobar and Zazula 2004; Kleine et al. 2007), the capacity of surface EMG recordings to discriminate motor-unit activity in a variety of conditions, as can be achieved with intramuscular EMG recordings, remains debatable.
Single motor-unit activity can be identified in surface EMG recordings only when the motor units are uniquely represented by their surface action potentials. This condition does not depend on the algorithm used to decompose the surface EMG or on the percentage of superimposed action potentials, but is related to the intrinsic information content of the surface EMG. If surface action potentials of individual motor units in a population all differ from each other, it should be possible to discriminate the discharge activity of each motor unit in the surface EMG recordings. The current study did not examine the capacity of specific decomposition algorithms (e.g., Kleine et al. 2007) to identify single motor units in the surface EMG, but rather we investigated the conditions that allow the discrimination of motor units on the basis of their surface EMG representations. When these conditions are not met, the action potentials of different motor units cannot be discriminated by any decomposition algorithm. The aim of the study was to investigate the relative proportion of motor units that are uniquely represented in the surface EMG detected with selected recording systems. The results provide the upper limit for detecting single motor units from the surface EMG. The ultimate success of any surface EMG decomposition procedure depends on the accuracy of the specific algorithm implemented, which is not addressed in this study.
METHODS
The study involved both computer simulations with a standardized model and experimental measurements performed on the abductor digiti minimi muscle. The simulations were based on an anatomical model that consisted of a cylindrical volume conductor (Farina et al. 2004a) with an anisotropic muscle layer, and isotropic bone, subcutaneous, and skin layers (Fig. 1). The model is similar to one used in a previous study (Keenan et al. 2005) and the parameters are reported in Table 1.
FIG. 1.
The anatomical model, recording systems, and representative motor unit action potentials. A: the model consisted of bone, muscle tissue, a subcutaneous layer, and skin. B: the location of the electrode grid over 3 sets of 3 muscle fibers (120 mm) with different locations for the innervation zones. The end plate is indicated as a short horizontal line across each muscle fiber. C: the filters used for obtaining the recordings. A recording was obtained by the summing, with weights indicated within the circles, of the signals detected by the electrodes that comprised the filter. For example, a bipolar recording involved subtracting the signals detected by the 2 electrodes (weights +1 and −1 for the 2 electrodes). A double differential recording was obtained by summing the signals detected by the 2 outer electrodes and subtracting twice the signal detected by the central electrode (weights: +1, −2, and +1). D: action potentials generated by a motor unit with 803 fibers that occupied about 40 mm2; the center of the motor-unit territory was located 0.2 mm in the transverse direction from the center of the muscle and 6.4 mm from the skin surface. The action potentials of the motor unit were recorded with 9 Laplacian filters (same configuration described in Fig. 2D).
TABLE 1.
Parameters of the cylindrical model
| Model Parameter | Value |
|---|---|
| Tissue conductivities, S/m | |
| Bone | 0.02 |
| Muscle (radial and transverse) | 0.1 |
| Muscle (longitudinal) | 0.5 |
| Subcutaneous tissue | 0.05 |
| Skin | 0.1 |
| Muscle properties | |
| Number of motor units | 200 |
| Number of muscle fibers | 119,634 |
| Number of fibers in a motor unit (range) | 15–1,500 |
| Muscle cross-sectional area | 598 mm2 |
| Average fiber length | 60 and 120 mm |
| Skin thickness | 1 mm |
| Subcutaneous tissue thickness | 1 and 5 mm |
| Bone (radius) | 20 mm |
| Tendon ending spread | 5 mm |
| Innervation zone spread | 5 mm |
| Electrodes | |
| Circular (diameter) | 1 mm |
| Interelectrode distance | 2.5 and 5 mm |
The muscle had an elliptical shape and was 30 mm in the transverse direction and 25.4 mm thick. The thickness of the subcutaneous layer was either 1 or 5 mm. The intracellular action potentials of the muscle fibers were derived from the analytical description developed by Rosenfalk (1969). Average fiber length was the same for all motor units and varied between 60 mm (short muscle) and 120 mm (long muscle). Fibers were parallel to each other and along the main axis of the cylindrical volume conductor (Keenan et al. 2005). The innervation zones of the motor units (n = 200) were located at −20 mm (−10 mm for the short muscle), 0 mm, or 20 mm (10 mm) from the center of the muscle (Fig. 1B). The locations of the end plates and tendon endings varied randomly (uniform distribution) over a range of 5 mm. The motor-unit territories were circular and distributed randomly throughout the muscle. The fibers of each motor unit were distributed with a density of 20 fibers/mm2 (Armstrong et al. 1988) and interdigitated with the fibers of many other units to yield a fiber density in the muscle of 200 fibers/mm2. When a portion of the motor-unit territory was constrained by the muscle boundary, the territory of the unit was modified to fit the muscle cross section (Keenan et al. 2005). Innervation numbers ranged from 15 to 1,500, based on an approximately 100-fold range of twitch forces (Elek et al. 1992), with an exponential distribution across the population (Fuglevand et al. 1993; Kernell 1992).
The motor units had a mean muscle fiber conduction velocity of 4.0 ± 0.35 m/s (range 3.2–5.0 m/s) (Farina et al. 2000; Troni et al. 1983), with the slowest conduction velocity assigned to the smallest motor unit (Andreassen and Arendt-Nielsen 1987). The surface-recorded motor-unit potential constituted the sum of the action potentials of the muscle fibers belonging to the motor unit. EMG signals were computed at 4,096 samples/s. Each simulated motor-unit population comprised 200 motor units. The action potentials for each motor unit were generated independently, without activating the other motor units. The action potentials generated by pairs of the 200 motor units in each population were compared.
Simulated recording systems
The simulated signals were detected with circular electrodes (diameter 1 mm), arranged in a grid with 11 rows and 11 columns (11 × 11 electrodes) with either 2.5 mm (short muscle) or 5 mm (long muscle) between electrodes in both the longitudinal and transverse directions (Fig. 1B). The center of the grid was either 7.5 mm (short muscle) or 15 mm (long muscle) distal to the center of the muscle in the longitudinal direction and placed over the center of the muscle in the transverse direction. Each simulated signal (channel) represented the filtered version of the electric potential generated by the motor unit at one of the locations over the skin. The simulated filters were monopolar, bipolar, double differential, and Laplacian (Fig. 1) for comparison with commonly used experimental measures (Hogrel 2003). In addition, the four-bipolar configuration (quadrupolar), developed by De Luca et al. (2006) to decompose the surface EMG, was also examined (Fig. 1C); this configuration provides four channels.
The channels were grouped in sets along the transverse and longitudinal directions to obtain different recording configurations with varying numbers of channels that could be used to discriminate motor-unit action potentials. Figure 2 shows examples of channel groupings for the Laplacian filter: one filter produces one channel of information and one recording (Fig. 2A), three filters aligned in a transverse direction provide three channels and three recordings (Fig. 2B), three filters arranged in a longitudinal direction also provide three channels and three recordings (Fig. 2C), and the combination of three filters in each of the transverse and longitudinal directions obtains nine channels and nine recordings (Fig. 2D). Similar configurations in the two directions were also created for 4 channels (2 rows and 2 columns), 25 channels (5 and 5), 49 channels (7 and 7), and 81 channels (9 and 9). These configurations will be indicated as arrays of 2 × 2, 3 × 3, 5 × 5, 7 × 7, and 9 × 9 channels. Each group of channels was arranged around the center of the 11 × 11 grid, as shown in Fig. 2, with a possible misalignment of one interelectrode distance in the case of even number of channels.
FIG. 2.
Schematic explanation of how the recording configurations were derived from the simulated grid of 11 rows and 11 columns of electrodes. The examples are for Laplacian filters in which each channel of information is obtained by the linear combination of the signals recorded from 5 electrodes (inset, top right corner). A: a single channel derived from the 5 filled electrodes that comprise the Laplacian filter; the open circles represent electrodes in the grid that were not used in this recording configuration. The center of the filter is indicated by the filled square. The action potential detected with this configuration is shown on the right for a representative motor unit. B: a configuration with 3 Laplacian filters arranged in the transverse direction. The 3 channels of output were obtained from 11 electrodes, with some electrodes shared by adjacent Laplacian filters, as it is commonly done in experiments (Hogrel 2003). C: 3 Laplacian filters arranged in the longitudinal direction. D: 9 Laplacian filters (3 transverse × 3 longitudinal) produce 9 signals, shown on the right for the same motor unit as for the other panels. The recording configurations for the other filters were obtained by substituting the Laplacian summation (with weights +1 for the 4 outer electrodes and −4 for the central electrode) with the summations shown in Fig. 1C. Configurations with 2 to 9 channels were simulated in both the transverse and longitudinal directions and ≤81 channels (9 × 9) in both directions. The motor-unit action potentials shown in this figure were obtained from a simulation with a subcutaneous layer that was 5 mm thick, the innervation zones located at −20 mm in the longitudinal direction from the center of the muscle, and fiber length of 120 mm. au, arbitrary units.
The simulations tested 22 recording configurations, which corresponded to one configuration of one channel (Fig. 2A), 8 configurations of channels arranged transversally (from 2 to 9 channels in the transverse direction, as shown for 3 channels in Fig. 2B), 8 configurations of channels arranged longitudinally (from 2 to 9 channels in the longitudinal direction, as shown for 3 channels in Fig. 2C), and 5 configurations of channels arranged in the two directions (2 × 2, 3 × 3, 5 × 5, 7 × 7, 9 × 9, as shown for 3 × 3 channels in Fig. 2D). Each of these configurations was tested with four filters (monopolar, bipolar, double differential, and Laplacian; Fig. 1C). Figure 2 shows the examples of 4 of the 22 configurations tested when the Laplacian filter was applied. In addition to these 22 recording configurations, the quadrupolar recording configuration was examined (Fig. 1C). This configuration provides 4 channels, as schematically shown in Fig. 1C, and was centered over the 11 × 11 grid. Thus each motor unit in the population generated an electrical distribution at the skin surface that was sampled by 22 different configurations of channels with 4 filters and the quadrupolar configuration. These arrangements provided 89 representations (22 configurations of channels × 4 filters plus the quadrupolar recording) of the electrical activity of each motor unit, and these were compared in their capacity to provide unique recordings of the 200 motor units in the simulated populations.
Simulation conditions and signal analysis
The simulated electrical activity of each motor unit consisted of a multichannel action potential derived from the set of action potentials recorded at each location. The multichannel action potentials of the 200 motor units were simulated as detected at the skin surface for each recording configuration and anatomical condition. The three anatomical conditions simulated were the length of the fibers (60 and 120 mm), the thickness of the subcutaneous tissue (1 and 5 mm), and the locations of the innervation zones. The latter condition involved all motor units being innervated in one of the three simulated locations and each motor unit being innervated randomly in one of the three locations.
Twenty populations of 200 motor units were simulated by randomly locating the motor units within the muscle tissue for each of the 12 anatomical conditions (2 fiber lengths, 2 subcutaneous thicknesses, 3 distributions of innervation zones). The action potentials of the 200 motor units in each of the 240 simulated populations of motor units (12 anatomies × 20 random distributions) were detected by 89 recording configurations (22 configurations of channels in the longitudinal, transverse, and combined directions × 4 filters, plus the quadrupolar configuration). Due to the amount of data generated for this study, only representative results are reported herein.
Pairs (n = 19,900) of simulated motor units in each population of 200 motor units were compared with every other motor unit in the population. The two multichannel action potentials for each pair of motor units were aligned in time by maximizing their cross-correlation function. The mean square difference was computed between the two aligned multichannel action potentials and was normalized (%) to the mean of the energies of the two action potentials. Pairs of action potentials with a mean square difference of <5% were considered identical and the two motor units were deemed indistinguishable. The 5% criterion was based on the variability observed experimentally in the energy of surface action potentials discharged by the same motor unit in the abductor digiti minimi, which was the muscle experimentally investigated in this study. The signals used to compute the variability were taken from Farina et al. (2004b); the variability in the shape of the surface action potential was about 6% of the energy of the signal for individual motor units. This resolution is necessary to distinguish the action potentials of a single motor unit from other units in the surface recording. Two motor units can be discriminated from each other if the difference in the action potential shapes is larger than the variability in the shape of the action potentials of each of the two motor units. The dependent variable was thus the number of motor units with multichannel action potentials that differed from all other motor units in the population. This number depended on the configuration used to record the action potentials. The approach does not depend on the number of superimposed action potentials in a recording; rather, the intent was to assess the upper limit in the number of motor units that could be discriminated based on their surface EMG representation.
In each condition, the results are reported as mean and SD for 20 motor-unit populations with the same recording system and anatomical characteristics, but with different random locations of the motor units within the muscle.
Experimental measurements
Experimental signals were recorded from the abductor digiti minimi muscle of eight healthy men (mean ± SD, 26.1 ± 3.2 yr) with 49 circular electrodes (diameter 1 mm) that were arranged in a grid (7 × 7 electrodes) with 2.5 mm between electrodes in both the longitudinal and the transverse directions. The study was conducted in accordance with the Declaration of Helsinki, approved by the local ethics committee (N-20070019), and written informed consent was obtained from all subjects prior to participating in the study. Ultrasound recordings (FFsonic UF-4000L, Fukuda Denshi) indicated that the thickness of the subcutaneous tissue over the abductor digiti minimi for the eight subjects was 1.7 ± 0.5 mm.
The electrode grid was placed over the distal portion of the muscle after light abrasion of the skin. The monopolar surface EMG signals were amplified (64-channel surface EMG amplifier, SEA 64, LISiN-OT Bioelettronica, Turin, Italy; −3 dB bandwidth 10–500 Hz), sampled at 2,048 Hz, and converted to digital form by a 12-bit A/D converter.
Motor units were identified from intramuscular recordings and the corresponding surface action potentials were extracted by spike-triggered averaging (Farina et al. 2002). To detect a relatively large number of motor units from each subject, the intramuscular EMG signals were recorded concurrently from two locations and during contractions at five target forces. Intramuscular EMG signals were recorded with two pairs of Teflon-coated stainless steel wires (diameter 0.1 mm; A-M Systems, Carlsborg, WA) inserted with 25-gauge hypodermic needles into two locations that were about 10 mm apart in the transverse direction in the proximal part of the muscle. The needles were inserted to a depth of a few millimeters below the muscle fascia and removed to leave the wire electrodes inside the muscle. The wires were cut to expose the cross section and the intramuscular EMG signals were differentially amplified (Counterpoint EMG, Dantec Medical, Skovlunde, Denmark), band-pass filtered (500 Hz to 5 kHz), sampled at 10 kHz, and stored after 12-bit A/D conversion. The gain of the intramuscular EMG amplifier was adjusted to maximize the amplitude of the action potentials at each contraction force. The position of the wires was slightly adjusted before the recordings and in a few cases the wires were reinserted when the signal-to-noise ratio was judged poor from visual inspection. Once the optimal location was determined, however, the position of the wires was not changed between contractions. Surface and intramuscular recordings were synchronized.
The fifth finger was fixed in a brace to record the force exerted during an isometric contraction of the muscle (Politecnico di Turin, Turin, Italy). The subjects performed three maximal voluntary contractions (MVCs) of the abductor digiti minimi with a 2-min rest between each MVC. The peak MVC force was used as the reference for the submaximal contraction forces. Five minutes after the MVCs, the subject performed five 30-s contractions at target forces of 2.5, 5, 7.5, 10, and 12.5% MVC force; there was a 5-min rest between each 30-s contraction.
The action potentials of the detected motor units were identified from the intramuscular recordings with a decomposition algorithm (McGill et al. 2005). Accuracy of the automatic part of this algorithm is >95% compared with expert manual decomposition (JR Florestal, PA Mathieu, and KC McGill, unpublished observations). Moreover, the algorithm includes a user interface for manually editing and verifying the results. The software displays a segment of the EMG signal, the templates of the action potentials of the identified motor units, the discharge patterns, and a close-up of the signal for resolving missed discharges and superimpositions. The automatic decomposition was checked by inspection of the identified discharge patterns. Full, regular patterns provided confidence that the decomposition was correct, whereas gaps, extra discharges, or uneven intervals indicated possible decomposition errors. To assist in identifying missed discharges, the program displays bars in the signal panel that indicate the expected discharge times of each motor unit. The intramuscular decomposition procedure has been validated (Florestal et al., unpublished observations).
The action potentials discharged by each identified motor unit were used to trigger an average from the multichannel surface EMG (Farina et al. 2002) to obtain isolated surface EMG action potentials for each motor unit, as in the simulations. Surface-recorded action potentials generated by the same motor unit at different target forces or concurrently at the two intramuscular locations were averaged. The identification of the same motor unit at different force levels was based on the shape of the intramuscular action potentials, which was the reference for the identification of a unique motor-unit action potential. Because the motor units were recorded at varying forces, however, it cannot be fully excluded that in some cases the same motor unit was included two or more times in the experimental population when detected at different forces. The shape of the intramuscular action potentials may have changed slightly with contraction force due to displacement of the wires. Thus the experimental population of motor units defined as unique on the basis of intramuscular recordings at different forces may have contained a few cases of nonunique motor units (see discussion).
The surface multichannel action potentials of the identified motor units, as obtained by spike-triggered averaging, were compared with the same procedures used with the simulated motor units and the same configurations of channels (Fig. 2), up to a maximum of five channels in each direction due to the smaller size of the electrode grid used experimentally with respect to the simulated grid. Results are reported for each subject as the percentage of identifiable motor units for representative recording configurations.
RESULTS
Motor units that were relatively similar in size but located in different parts of the muscle often projected similar motor-unit potentials onto surface recordings, especially when few channels were used to obtain the recording. Figure 3 shows a group of eight simulated motor-unit action potentials, as recorded by a single bipolar channel, that were indistinguishable from each other with this recording configuration.
FIG. 3.
Simulated bipolar recordings of motor-unit potentials in the long muscle (120-mm fiber length) from one location in the 11 × 11 electrode grid (central location). From the population of 200 motor units (circles in the left traces), a group of 8 (filled black circles in the left traces) was selected. These 8 motor units were located in different parts of the muscle and had slightly different sizes. A recording with one bipolar channel detected the potentials generated by the eight motor units, and these could not be distinguished from one another based on the criterion of 5% threshold in energy. In this simulation, the subcutaneous layer was 5 mm thick and the innervation zone was located at −20 mm in the longitudinal direction from the center of the long muscle. au: arbitrary units.
The percentage of motor units that could be identified depended on the number and distribution of channels in the different recording configurations. For example, the percentage of motor units that could be identified with one channel for the long muscle was <5% of the population with bipolar filtering and <20% with Laplacian filtering (Fig. 4 A). An increase in the number of channels in the longitudinal direction produced only a modest increase in the percentage of identifiable motor units when the motor units were all innervated in the same location (Fig. 4A). There was a marginal improvement in discrimination when the motor-unit potentials were compared with a nine-channel Laplacian system in the longitudinal direction. In contrast, an increase in the number of channels in the transverse direction dramatically improved the discrimination of motor-unit potentials. Approximately 50% of the 200 simulated motor units could be distinguished from the configuration with nine channels aligned in the transverse direction and bipolar filtering. This percentage further increased to 80% when Laplacian filtering was used (Fig. 4A).
FIG. 4.
The percentage of simulated motor units that could be identified in selected anatomical conditions with different recording configurations in the long muscle (fiber length: 120 mm). A: the influence of the number of channels on the estimates obtained with bipolar and Laplacian filtering for configurations with channels in both transverse and longitudinal directions when the innervation zone for all motor units was located at −20 mm in the longitudinal direction from the muscle center. The configurations with Laplacian filtering and 3 channels, for example, were obtained as described in Fig. 2, B and C, respectively. The configuration with one channel corresponds to that described for the Laplacian filtering in Fig. 2A. The subcutaneous layer was 1 mm thick. B: the estimates obtained from bipolar and Laplacian filtering with configurations in both directions (transverse and longitudinal), when the thickness of the subcutaneous layer was either 1 or 5 mm. The configuration indicated with 3 × 3, for example, is that described in Fig. 2D for the 9 Laplacian filters. C and D: the same conditions as shown in A and B, respectively, but with random variation in the location of the innervation zone for each motor unit between one of 3 positions: −20, 0, or 20 mm in the longitudinal direction from the center of the muscle. Results are reported as mean and SD for the 20 simulated populations of motor units.
Similar results were obtained for the configurations consisting of channels in the two directions (Fig. 4B). The thickness of the subcutaneous layer reduced the number of motor units that could be identified from bipolar recordings, but it had a negligible effect on the estimates obtained with Laplacian filtering for the same number of channels (Fig. 4B). The smaller sensitivity of Laplacian recordings to the thickness of the subcutaneous layer was due to the greater spatial selectivity of this recording system with respect to the other systems analyzed (Reucher et al. 1987). The percentages of identifiable motor units in the long muscle for all recording configurations when the motor-unit populations had the same location for the center of the innervation zone and 5 mm of subcutaneous tissue are reported in Table 2. Similar results were obtained for populations of motor units with different innervation-zone locations and only 1 mm of subcutaneous tissue. However, thinner subcutaneous tissue enabled the identification of larger percentages of motor units for monopolar, bipolar, and double differential filtering. For example, the configuration of 81 channels arranged in the two directions (9 × 9 channels) with bipolar filtering could discriminate 57.0 ± 3.6% of the motor units when the subcutaneous tissue was 1 mm thick in the same conditions as in Table 2.
TABLE 2.
Percentage of identifiable motor units detected in the surface EMG with monopolar, bipolar, double differential, and Laplacian filtering, with channels placed transversely across the muscle, along its length, and in both directions
| Number of Channels | Monopolar | Bipolar | Double Differential | Laplacian | Quadrupolar | |
|---|---|---|---|---|---|---|
| Transverse | 1 | 3.4 ± 1.2 | 1.3 ± 0.8 | 1.3 ± 0.7 | 24.7 ± 2.7 | — |
| 2 | 10.5 ± 2.3 | 8.6 ± 1.6 | 11.7 ± 2.0 | 53.6 ± 3.7 | — | |
| 2 | 10.5 ± 2.3 | 8.6 ± 1.6 | 11.7 ± 2.0 | 53.6 ± 3.7 | — | |
| 3 | 17.2 ± 2.9 | 15.6 ± 2.1 | 19.4 ± 3.0 | 64.0 ± 3.0 | — | |
| 4 | 21.7 ± 3.3 | 19.7 ± 2.5 | 25.1 ± 4.1 | 70.3 ± 3.7 | — | |
| 5 | 26.4 ± 3.3 | 24.0 ± 2.9 | 30.6 ± 4.0 | 73.9 ± 3.7 | — | |
| 6 | 29.4 ± 4.1 | 27.6 ± 2.5 | 35.0 ± 3.7 | 76.9 ± 3.7 | — | |
| 7 | 31.8 ± 4.1 | 30.5 ± 3.7 | 39.5 ± 3.9 | 78.3 ± 3.5 | — | |
| 8 | 33.9 ± 3.9 | 32.1 ± 3.4 | 42.0 ± 3.6 | 79.4 ± 3.6 | — | |
| 9 | 35.6 ± 3.8 | 35.1 ± 3.0 | 43.7 ± 2.7 | 80.4 ± 3.4 | — | |
| Longitudinal | 2 | 3.4 ± 1.2 | 1.5 ± 0.9 | 1.4 ± 0.6 | 25.0 ± 2.9 | — |
| 3 | 3.7 ± 1.3 | 1.7 ± 0.9 | 1.7 ± 0.8 | 26.0 ± 2.9 | — | |
| 4 | 3.8 ± 1.3 | 2.0 ± 0.9 | 2.4 ± 1.0 | 26.6 ± 3.1 | — | |
| 5 | 3.9 ± 1.3 | 2.4 ± 1.0 | 2.9 ± 1.2 | 28.2 ± 3.0 | — | |
| 6 | 4.2 ± 1.4 | 2.9 ± 1.1 | 4.0 ± 1.3 | 29.1 ± 2.8 | — | |
| 7 | 4.1 ± 1.4 | 3.2 ± 1.2 | 4.9 ± 1.6 | 31.4 ± 3.3 | — | |
| 8 | 4.4 ± 1.5 | 3.8 ± 1.3 | 6.1 ± 1.7 | 33.0 ± 3.5 | — | |
| 9 | 4.4 ± 1.3 | 4.3 ± 1.4 | 7.1 ± 2.0 | 34.7 ± 3.8 | — | |
| Both directions | 2 × 2 | 11.5 ± 2.7 | 9.9 ± 1.8 | 11.7 ± 1.6 | 56.3 ± 3.6 | 38.9 ± 3.4 |
| 3 × 3 | 17.4 ± 3.2 | 17.4 ± 2.1 | 21.5 ± 2.8 | 65.9 ± 3.3 | — | |
| 5 × 5 | 25.4 ± 3.1 | 27.2 ± 3.6 | 35.7 ± 3.5 | 76.0 ± 3.7 | — | |
| 7 × 7 | 30.4 ± 3.9 | 35.5 ± 3.6 | 46.5 ± 3.6 | 81.1 ± 3.6 | — | |
| 9 × 9 | 32.7 ± 4.0 | 41.4 ± 3.9 | 52.5 ± 3.9 | 83.8 ± 4.0 | — |
Values are means ± SD for 20 populations of motor units that were distributed randomly within the muscle. The subcutaneous layer was 5mm thick and the innervation zone was located at −20 mm from the center of the muscle.
Results on the short muscle and 2.5-mm distance between electrodes were in agreement with those for the long muscle. With a subcutaneous thickness of 1 mm (5 mm), for example, the configuration with a single bipolar channel discriminated 3.5 ± 1.1% (1.7 ± 1.1%) of the motor units, whereas the configuration with 9 channels in the longitudinal direction with the same filtering identified 6.6 ± 1.8% (3.1 ± 1.0%), the configuration with 9 bipolar channels in the transverse direction distinguished 46.7 ± 3.1% (27.3 ± 3.4%), and the configuration with 81 bipolar channels in the two directions (9 × 9 channels) revealed 47.8 ± 2.9% (27.5 ± 3.4%) of the motor units. As for the long muscle, the configuration with 81 channels (9 × 9) with Laplacian filtering could discriminate almost the entire population of motor units in the short muscle (90.4 ± 1.4 and 90.2 ± 1.4% for subcutaneous layers of 1 and 5 mm, respectively). Thus the general trend for the results was identical for the long and short muscles and for the two interelectrode distances.
Variation in the location of the center of the innervation zone for each motor unit improved the discrimination capacity for some recording configurations. The locations of the innervation zone were randomly varied among three positions: −20 mm (−10 mm for the short muscle), 0 mm, and 20 mm (10 mm for the short muscle) from the muscle center (Fig. 1). The greatest improvement in performance occurred for the estimates derived from bipolar filtering (Fig. 4, C and D). When the subcutaneous layer was 1 mm thick and the innervation zone locations were scattered in the long muscle, the configuration of 81 channels (9 × 9) with bipolar filtering identified a similar percentage of motor units as the same configuration with Laplacian filtering (Fig. 4D). Similar results were obtained for the short muscle, where the configuration of 9 × 9 bipolar channels identified 76.9 ± 4.0% of the motor units when the innervation zones were scattered and the subcutaneous layer was 1 mm thick.
Because low-threshold units are usually analyzed in experimental studies, the percentage of identifiable motor units was further analyzed as a function of motor-unit territory. There was a tendency for motor units with small territories to be better identified than larger units (results are shown for the long muscle in Fig. 5A). However, this effect was absent for motor units that innervated >100 fibers. There was no electrode configuration that was able to discriminate all motor units in a specific range of recruitment thresholds (e.g., low-threshold motor units) when using few channels. The motor units that could not be distinguished tended to have a similar size and location (Fig. 3); these motor units would be activated at similar forces during voluntary contractions. Results on the short muscle were comparable with those for the long muscle.
FIG. 5.
The influence of innervation number and motor-unit location on the number of identifiable motor units in the simulated long muscle (fiber length: 120 mm). A: the proportion of unique motor units was determined as a percentage of the number of motor units with similar innervation numbers. Bin width was 50 fibers. B: the proportion of unique motor units is shown as a percentage of the number of motor units located at each depth. Bin width was 2 mm. C: the proportion of unique motor units as a percentage of the number of motor units closer to the electrodes than the specified distance. All motor units were innervated at −20-mm distance from the muscle center and the subcutaneous layer was 1 mm. The results are reported as means for 20 simulated populations of motor units.
Because superficial motor units tend to present larger action potentials than deeper motor units, the analysis of motor units from the surface EMG is biased toward the most superficial. Therefore the percentage of identifiable motor units was also analyzed as a function of the distance from the electrodes to the center of the motor-unit territory. With the exception of the Laplacian filtering, the percentage of identifiable motor units tended to be larger for closer motor units (results for the long muscle in Fig. 6) when the percentage was expressed relative to the number of motor units within specified ranges of distance (Fig. 5B) or relative to the number of motor units within a specified distance from the electrodes (Fig. 5C). However, the effect of distance was modest for all recording configurations (Fig. 5C). Figure 6 shows the location of the motor units that were uniquely represented by various surface EMG recording configurations in one of the anatomical conditions.
FIG. 6.
Representation of identifiable motor units in a simulated population for the long muscle (fiber length: 120 mm). A: the territories of all simulated motor units. From this population, the territories of motor units that were identifiable are shown for monopolar (B, F), bipolar (C, G), double differential (D, H), and Laplacian (E, I) filtering in the case of 4 (2 × 2) channels (B, C, D, E) and 81 (9 × 9) channels (F, G, H, I). The proportion of identifiable motor units is also indicated as a percentage of the entire population. The subcutaneous layer was 1 mm thick and all motor units were innervated at −20 mm from the muscle center.
Figure 7 shows the percentage of identifiable motor units as a combined function of the number of fibers and distance from the skin for two configurations of channels, when the subcutaneous tissue was 5 mm thick. There was a trend for large and distant motor units to be indistinguishable. When the number of channels increased from 4 (2 × 2 channel configuration in Fig. 7, A and B) to 81 (9 × 9 channel configuration in Fig. 7, D and E), the percentage of unique motor units increased. The quadrupolar configuration (described in Fig. 1C), which provides 4 channels, produced results similar to those of the configuration with 4 channels (2 × 2) using Laplacian filtering (compare Fig. 7, C to B). Laplacian and bipolar filtering had similar capacities to identify superficial (<15 mm) motor units when the configuration with 81 channels (9 × 9) was used in the discrimination (Fig. 7, D and E). Similar trends were observed when the subcutaneous tissue was 1 mm thick, but greater numbers of motor units were identifiable and the results were more similar for the bipolar and Laplacian filtering. The influence of distance and number of fibers on the percentage of identifiable motor units shown for the long muscle was similar for the short muscle.
FIG. 7.
The combined influence of innervation number and motor-unit location on the number of identifiable motor units for the simulated long muscle (fiber length: 120 mm). The proportion of motor units of different sizes that could be identified at the different locations was determined as a percentage of the number of motor units in bins of 50 fibers and of 2-mm distances from the skin. The percentage of identifiable motor units is shown for configurations of 4 channels (2 × 2) (A, B) and 81 channels (9 × 9) (D, E) using bipolar filtering (A, D), Laplacian filtering (B, E), and the quadrupolar system (C). The subcutaneous layer was 5 mm thick and all motor units were innervated at −20 mm from the muscle center.
Spike-triggered averaging of experimental recordings from the abductor digiti minimi was used to determine the shapes of the surface-detected, multichannel action potentials during voluntary contractions. The number of triggers used to average the potentials of 189 motor units from eight subjects was 238 ± 75.3. Pairs of motor unit action potentials were often identical when using one channel and bipolar filtering, but differed when compared with a configuration of 25 channels (5 × 5 channels) (Fig. 8), as with the results obtained in the simulations. For example, Fig. 8A shows one motor unit (top set of traces) with surface action potentials that increased in amplitude toward the left column of the grid, whereas the amplitude of the action potentials for a second motor unit (bottom set of traces) was maximum in the middle column; thus the two units were clearly distinguishable in the configuration with 25 channels arranged in the two directions. In contrast, the recordings obtained from the 5 channels in the longitudinal direction (middle column of the grid) were similar for the two motor units. Comparable results were observed for the pairs of motor units from the eight subjects investigated (Table 3). The percentage of distinguishable motor units across the eight subjects was 33.7 ± 14.2 and 39.7 ± 11.1% when using one single bipolar or Laplacian channel, respectively; 52.8 ± 11.2 and 64.0 ± 7.6% with configurations of 3 × 3 channels with bipolar and Laplacian filtering; and 80.5 ± 9.4 and 86.5 ± 11.1% with configurations of 5 × 5 channels.
FIG. 8.
Multichannel surface recordings of motor-unit action potentials measured experimentally from the abductor digiti minimi muscle of one subject. A: the multichannel action potentials of 2 motor units detected at 5% MVC have been estimated by spike-triggered averaging. The 2 motor units could be distinguished with a configuration of 25 channels (5 × 5) with bipolar filtering, but not with one bipolar channel (enclosed in the dashed square). The 2 motor-unit potentials are superimposed in the solid square. Similarly, the 2 motor-unit potentials were not distinguishable when using the entire middle column (5 channels arranged in the longitudinal direction with bipolar filtering). B: same as in A for another pair of motor units detected at 2.5% maximal voluntary contraction (MVC) from the same muscle. The intramuscular action potentials used for spike-triggered averaging are shown on the left of the surface recordings.
TABLE 3.
Number of identified motor units that were considered unique
| Number of Motor Units | Motor Unit Pairs | Number of Channels | Bipolar | Laplacian | |
|---|---|---|---|---|---|
| Subject 1 | 23 | 253 | 1 | 7 (30.4%) | 8 (34.8%) |
| 3 × 3 | 13 (56.5%) | 15 (65.2%) | |||
| 5 × 5 | 18 (78.3%) | 21 (91.3%) | |||
| Subject 2 | 27 | 351 | 1 | 6 (22.2%) | 9 (33.3%) |
| 3 × 3 | 12 (44.4%) | 18 (66.7%) | |||
| 5 × 5 | 21 (77.8%) | 22 (81.5%) | |||
| Subject 3 | 18 | 153 | 1 | 2 (11.1%) | 4 (22.2%) |
| 3 × 3 | 6 (33.3%) | 10 (55.6%) | |||
| 5 × 5 | 11 (61.1%) | 13 (72.2%) | |||
| Subject 4 | 15 | 105 | 1 | 6 (40.0%) | 6 (40.0%) |
| 3 × 3 | 8 (53.3%) | 9 (60.0%) | |||
| 5 × 5 | 13 (86.7%) | 15 (100%) | |||
| Subject 5 | 25 | 300 | 1 | 13 (52.0%) | 15 (60.0%) |
| 3 × 3 | 16 (64.0%) | 18 (72.0%) | |||
| 5 × 5 | 23 (92.0%) | 23 (92.0%) | |||
| Subject 6 | 25 | 300 | 1 | 12 (48.0%) | 12 (48.0%) |
| 3 × 3 | 12 (48.0%) | 14 (56.0%) | |||
| 5 × 5 | 20 (80.0%) | 18 (72.0%) | |||
| Subject 7 | 30 | 435 | 1 | 7 (23.3%) | 11 (36.7%) |
| 3 × 3 | 16 (53.3%) | 18 (60.0%) | |||
| 5 × 5 | 24 (80.0%) | 25 (83.3%) | |||
| Subject 8 | 26 | 325 | 1 | 11 (42.3%) | 11 (42.3%) |
| 3 × 3 | 18 (69.2%) | 20 (76.9%) | |||
| 5 × 5 | 23 (88.5%) | 26 (91.3%) |
Motor units were considered unique on the basis of their intramuscular action potential shapes, number of motor unit pairs compared, number of channels used for comparison, and number (percentage) of motor units that could be discriminated using 1, 3 × 3, and 5 × 5 bipolar and Laplacian systems for the eight subjects investigated. The recording configurations were obtained as explained in Fig. 2 for the simulations.
DISCUSSION
This study examined the capacity of selected recording configurations to discriminate the action potentials of single motor units in surface EMG recordings. The number of motor-unit potentials that could be identified in the surface recordings depended on the number of channels used in the discrimination. The proportion of the motor-unit population that could be identified was small when few surface EMG channels were used to detect the simulated signals and experimental recordings, but it increased to most of the population when multiple channels (9 × 9 and 5 × 5 channels, respectively) were used in the analysis. The simulations also indicated that the discrimination capacity depended on the anatomical characteristics of the system.
Due to the low-pass filtering effects of the tissues interposed between the active muscle fibers and the electrodes, surface-detected action potentials for different motor units can be similar and thus indistinguishable (Fig. 3). Under some conditions, it is possible to identify motor-unit potentials in the surface EMG (De Luca et al. 2006; Gazzoni et al. 2004; Kleine et al. 2007). However, there has been no theoretical evaluation of the use of surface EMG recordings to study motor-unit activity. The approach used in the present study was to determine the conditions that enabled the appearance of unique motor-unit potentials in the simulated surface EMG recording and to verify the outcomes with experimental recordings. The analysis does not depend on the decomposition method because the signal cannot be decomposed if it does not contain motor-unit potentials with unique shapes. The findings in the current study thus indicate the upper limit in number of units that can be discriminated from the surface EMG. In practice, the number of motor units and action potential discharges that can be identified will depend on the ability of the decomposition algorithm to detect and classify action potentials and on the degree of complexity of the signal (e.g., percentage of superimposed action potentials), which has not been addressed in this study.
The simulation and experimental results show that when only a few channels of surface EMG recordings are used in the discrimination, it is not possible to discriminate surface motor-unit potentials. The information content of a few surface EMG channels is thus not sufficient for investigating the activity of single motor units. On average, only 33.7% of the low-threshold motor units recorded experimentally could be distinguished with one bipolar channel. In some special cases, decomposition of the surface EMG from few recorded channels is possible (De Luca et al. 2006; Hogrel 2003), but the results do not generalize and the methods have only limited applicability. The poor discrimination performance of a few channels of recordings applies to motor units with different sizes and at different distances from the recording electrodes. For example, when limiting the motor units of interest to those within 10 mm of the electrodes (because the others may have action potentials below the noise level), the percentage of simulated motor units that can be discriminated with a single bipolar or Laplacian channel is <20% (Fig. 5C), which is consistent with the percentage of identifiable motor units observed in the experimental recordings. Furthermore, <60% of the simulated motor units could be discriminated under the same conditions with the system of four channels proposed by De Luca et al. (2006) (Fig. 5C). Many of the motor units that cannot be identified are superficial and small (Fig. 7), which are those most likely to be recorded during voluntary contractions in motor-unit studies. These indistinguishable motor units cannot be discriminated by any decomposition algorithm because they appear as a single source.
The number of identifiable motor units in the simulated signals increased substantially when there was an increase in the number of channels used in the discrimination process. A similar result was obtained with the experimental measurements, which showed that about 80% of the detected motor units could be distinguished with a multichannel system. In experimental conditions, the shape of the intramuscular action potentials was used to ensure that the motor units detected for each subject represented a population of unique units. However, the experimental recordings were performed at various forces to identify lower-threshold motor units, which are usually masked by larger action potentials at higher forces. At each force level, all motor units that contributed to the signal were identified by automatic and interactive decomposition. The motor units detected at different forces were compared and merged when the same motor unit was identified. Although this procedure involved visual inspection of the shapes of the intramuscular action potentials and computation of their mean square errors, it cannot be excluded that some units were included more than once in the experimental population due to a change in the shape of their intramuscular action potential shapes. The number of motor units reported in Table 3 may overestimate the actual number of individual motor units detected experimentally.
Although the averaging was performed with all available triggers, it is possible that the variability in the shape of the averaged surface potentials was >5%, which was the level assumed in the simulations. The larger variability in surface potentials may have decreased the capacity to discriminate among motor units. The potential overestimation in the number of unique motor units based on intramuscular recordings and the potential larger variability in surface EMG action potential shapes due to spike-triggered averaging with a finite number of triggers indicate that the percentage of motor units that could be discriminated from experimental recordings as reported in Table 3 may have been underestimated. Thus the data in Table 3 likely provide a conservative estimate of the number of motor unit action potentials that can be discriminated experimentally based on surface EMG. These problems would have a similar influence on all channel configurations and thus would not alter the general conclusions.
The results indicate that large electrode grids are necessary to discriminate a high proportion of the motor-unit population from noninvasive surface EMG recordings (Kleine et al. 2007). Contrary to intramuscular recordings, it is relatively easy to increase the number of recording locations when applying surface electrodes. High-density surface EMG systems have been developed in several research laboratories (Blok et al. 2002; Gazzoni et al. 2004; Holtermann et al. 2005; Kleine et al. 2007; Lapatki et al. 2006; Madeleine et al. 2006) with construction techniques that allow for easy placement of the grid over a muscle of interest. Although the current study shows that motor units can be discriminated in surface EMG recordings under some conditions, the work did not consider the development of algorithms that could detect and discriminate motor unit action potentials in the interference EMG.
The ability to distinguish motor unit action potentials was influenced by the recording configuration and the anatomical characteristics of the system. The discrimination capacity of Laplacian filtering is greater than that of monopolar recordings (Table 2) due to the effect that differentiation has on enhancing the differences between action potentials generated by motor units at different locations in the muscle. The ability to identify individual motor-unit potentials was also influenced by selected anatomical conditions, such as the thickness of the subcutaneous layer and the location of the innervation zones. Although the simulations reported in this study varied only a few anatomical parameters, similar trends were also observed for other simulation conditions (data not reported). For example, the conclusions were not influenced by the muscle cross-sectional area or number of motor units. Similarly, the uniform distribution of innervation number across the motor-unit pool, instead of the exponential distribution reported in this study, led to almost identical results. Thus although the present report focuses on only a subset of results, all conclusions were confirmed in a larger set of simulations.
Although the simulation results are limited by the relatively simple geometry of the volume conductor, more complex geometries would not have substantially changed the results. For example, simulation of fiber pennation (Mesin and Farina 2004) and tissue inhomogeneities (Mesin and Farina 2005) had an effect on action potential amplitude but not on the shape of the motor unit action potentials. The influences of fiber pennation and tissue inhomogeneities on action potential amplitude were similar across motor units, spatial filters, and multichannel configurations (Mesin and Farina 2004, 2005).
In summary, the simulation data and the experimental recordings indicate that relatively few motor units are distinguishable when only few channels of surface EMG signals are available for discrimination. In contrast, currently available multichannel, surface EMG recordings can discriminate a high proportion of the motor units in a simulated population and thus theoretically permit single motor-unit analysis. The extraction of motor-unit activity from surface EMG signals thus requires the use of electrode grids and not simpler detection systems that comprise only a few channels of recordings.
GRANTS
This work was supported by the Danish Technical Research Council Project “Centre for Neuroengineering” under Contract 26-04-0100 to D. Farina and by National Institute of Neurological Disorders and Stroke Grant NS-43275 to R. M. Enoka.
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
REFERENCES
- Adrian 1929.Adrian ED, Bronk DW. The discharge of impulses in motor nerve fibres: Part II. The frequency of discharge in reflex and voluntary contractions. J Physiol 67: 119–151, 1929. [PMC free article] [PubMed] [Google Scholar]
- Andreassen 1987.Andreassen S, Arendt-Nielsen L. Muscle fibre conduction velocity in motor units of the human anterior tibial muscle: a new size principle parameter. J Physiol 391: 561–571, 1987. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Armstrong 1988.Armstrong JB, Rose PK, Vanner S, Bakker GJ, Richmond FJ. Compartmentalization of motor units in the cat neck muscle, biventer cervicis. J Neurophysiol 60: 30–45, 1988. [DOI] [PubMed] [Google Scholar]
- Blok 2002.Blok JH, van Dijk JP, Drost G, Zwarts MJ, Stegeman DF. A high-density multichannel surface electromyography system for the characterization of single motor units. Rev Sci Instrum 73: 1887–1897, 2002. [Google Scholar]
- De Luca 2006.De Luca CJ, Adam A, Wotiz R, Gilmore LD, Nawab SH. Decomposition of surface EMG signals. J Neurophysiol 96: 1646–1657, 2006. [DOI] [PubMed] [Google Scholar]
- Dimitrov 1974.Dimitrov GV, Dimitrova NA. Extracellular potential field of a single striated muscle fibre immersed in anisotropic volume conductor. Electromyogr Clin Neurophysiol 14: 423–436, 1974. [PubMed] [Google Scholar]
- Elek 1992.Elek JM, Kossev A, Dengler R, Schubert M, Wohlfahrt K, Wolf W. Parameters of human motor unit twitches obtained by intramuscular microstimulation. Neuromuscul Disord 2: 261–267, 1992. [DOI] [PubMed] [Google Scholar]
- Farina 2002.Farina D, Arendt-Nielsen L, Merletti R, Graven-Nielsen T. Assessment of single motor unit conduction velocity during sustained contractions of the tibialis anterior muscle with advanced spike triggered averaging. J Neurosci Methods 115: 1–12, 2002. [DOI] [PubMed] [Google Scholar]
- Farina 2000.Farina D, Fortunato E, Merletti R. Noninvasive estimation of motor unit conduction velocity distribution using linear electrode arrays. IEEE Trans Biomed Eng 47: 380–388, 2000. [DOI] [PubMed] [Google Scholar]
- Farina 2004b.Farina D, Gazzoni M, Camelia F. Low-threshold motor unit membrane properties vary with contraction intensity during sustained activation with surface EMG visual feedback. J Appl Physiol 96: 1505–1515, 2004b. [DOI] [PubMed] [Google Scholar]
- Farina 2004a.Farina D, Mesin L, Martina S, Merletti R. A surface EMG generation model with multilayer cylindrical description of the volume conductor. IEEE Trans Biomed Eng 51: 415–426, 2004a. [DOI] [PubMed] [Google Scholar]
- Fuglevand 1993.Fuglevand AJ, Winter DA, Patla AE. Models of recruitment and rate coding organization in motor-unit pools. J Neurophysiol 70: 2470–2488, 1993. [DOI] [PubMed] [Google Scholar]
- Gazzoni 2004.Gazzoni M, Farina D, Merletti R. A new method for the extraction and classification of single motor unit action potentials from surface EMG signals. J Neurosci Methods 136: 165–177, 2004. [DOI] [PubMed] [Google Scholar]
- Hogrel 2003.Hogrel JY Use of surface EMG for studying motor unit recruitment during isometric linear force ramp. J Electromyogr Kinesiol 13: 417–423, 2003. [DOI] [PubMed] [Google Scholar]
- Holobar 2004.Holobar A, Zazula D. Correlation-based decomposition of surface electromyograms at low contraction forces. Med Biol Eng Comput 42: 487–495, 2004. [DOI] [PubMed] [Google Scholar]
- Holobar 2006.Holobar A, Zazula D, Gazzoni M, Merletti R, Farina D. Noninvasive analysis of motor unit discharge patterns in isometric force-varying contractions. Proc XVI Congr Int Soc Electrophysiol Kinesiol, June 28–July 1, Torino, Italy, 2006, p. 12.
- Holtermann 2005.Holtermann A, Roeleveld K, Karlsson JS. Inhomogeneities in muscle activation reveal motor unit recruitment. J Electromyogr Kinesiol 15: 131–137, 2005. [DOI] [PubMed] [Google Scholar]
- Keenan 2005.Keenan KG, Farina D, Maluf KS, Merletti R, Enoka RM. Influence of amplitude cancellation on the simulated surface electromyogram. J Appl Physiol 98: 120–131, 2005. [DOI] [PubMed] [Google Scholar]
- Kernell 1992.Kernell D Organized variability in the neuromuscular system: a survey of task-related adaptations. Arch Ital Biol 130: 19–66, 1992. [PubMed] [Google Scholar]
- Kleine 2000.Kleine BU, Blok JH, Oostenveld R, Praamstra P, Stegeman DF. Magnetic stimulation-induced modulations of motor unit firings extracted from multi-channel surface EMG. Muscle Nerve 23: 1005–1015, 2000. [DOI] [PubMed] [Google Scholar]
- Kleine 2007.Kleine BU, van Dijk JP, Lapatki BG, Zwarts MJ, Stegeman DF. Using two-dimensional spatial information in decomposition of surface EMG signals. J Electromyogr Kinesiol 17: 535–548, 2007. [DOI] [PubMed] [Google Scholar]
- Kleine 2008.Kleine BU, van Dijk JP, Zwarts MJ, Stegeman DF. Inter-operator agreement in decomposition of motor unit firings from high-density surface EMG. J Electromyogr Kinesiol 18: 652–661, 2008. [DOI] [PubMed] [Google Scholar]
- Lapatki 2006.Lapatki BG, Oostenveld R, Van Dijk JP, Jonas IE, Zwarts MJ, Stegeman DF. Topographical characteristics of motor units of the lower facial musculature revealed by means of high-density surface EMG. J Neurophysiol 95: 342–354, 2006. [DOI] [PubMed] [Google Scholar]
- Madeleine 2006.Madeleine P, Leclerc F, Arendt-Nielsen L, Ravier P, Farina D. Experimental muscle pain changes the spatial distribution of upper trapezius muscle activity during sustained contraction. Clin Neurophysiol 117: 2436–2445, 2006. [DOI] [PubMed] [Google Scholar]
- Masuda 1985.Masuda T, Miyano H, Sadoyama T. The position of innervation zones in the biceps brachii investigated by surface electromyography. IEEE Trans Biomed Eng 32: 36–42, 1985. [DOI] [PubMed] [Google Scholar]
- McGill 2005.McGill KC, Lateva ZC, Marateb HR. EMGLAB: an interactive EMG decomposition program. J Neurosci Methods 149: 121–133, 2005. [DOI] [PubMed] [Google Scholar]
- Merletti 1999.Merletti R, Lo Conte L, Avignone E, Guglielminotti P. Modeling of surface myoelectric signals—Part I: Model implementation. IEEE Trans Biomed Eng 46: 810–820, 1999. [DOI] [PubMed] [Google Scholar]
- Mesin 2004.Mesin L, Farina D. Simulation of surface EMG signals generated by muscle tissues with inhomogeneity due to fiber pinnation. IEEE Trans Biomed Eng 51: 1521–1529, 2004. [DOI] [PubMed] [Google Scholar]
- Mesin 2005.Mesin L, Farina D. A model for surface EMG generation in volume conductors with spherical inhomogeneities. IEEE Trans Biomed Eng 52: 1984–1993, 2005. [DOI] [PubMed] [Google Scholar]
- Reucher 1987.Reucher H, Silny J, Rau G. Spatial filtering of noninvasive multielectrode EMG: Part II—Filter performance in theory and modeling. IEEE Trans Biomed Eng 34: 106–113, 1987. [DOI] [PubMed] [Google Scholar]
- Rosenfalck 1969.Rosenfalck P Intra- and extracellular potential fields of active nerve and muscle fibers. Acta Physiol Scand Suppl 47: 239–246, 1969. [PubMed] [Google Scholar]
- Stegeman 2000.Stegeman DF, Blok JH, Hermens HJ, Roeleveld K. Surface EMG models: properties and applications. J Electromyogr Kinesiol 10: 313–326, 2000. [DOI] [PubMed] [Google Scholar]
- Troni 1983.Troni W, Cantello R, Rainero I. Conduction velocity along human muscle fibers in situ. Neurology 33: 1453–1459, 1983. [DOI] [PubMed] [Google Scholar]
- Zazula 2005.Zazula D, Holobar A. An approach to surface EMG decomposition based on higher-order cumulants. Comput Methods Programs Biomed Suppl 1: S51–S60, 2005. [DOI] [PubMed] [Google Scholar]
- Zwarts 2004.Zwarts MJ, Lapatki BG, Kleine BU, Stegeman DF. Surface EMG: how far can you go? Suppl Clin Neurophysiol 57: 111–119, 2004. [DOI] [PubMed] [Google Scholar]