Interpretation of the surface EMG is complicated by many factors that influence this signal (8). Experimental paradigms are often simplified to reduce the number of variables relevant for the interpretation. Many laboratory studies are based on sustained, constant-force contractions in one posture (hereafter termed "static" contractions), usually of high effort and short duration. In most applied fields, however, the static contraction paradigm is prohibitive, and the EMG must generally be analyzed during contractions in which both force and posture may vary (hereafter termed "dynamic" contractions). The difficulties in interpretation of the surface EMG in static contractions are amplified in the dynamic case. In dynamic tasks, much less is known about either the appropriate methods to analyze the EMG or the relations between signal features and the underlying physiological mechanisms. This issue is worsened by the lack of mathematical models to accurately describe generation of the EMG signal during movement. Nevertheless, the importance of dynamic surface EMG has led to its widespread use in movement analysis with many relevant applications.
This brief review discusses some of the techniques developed for the analysis of surface EMG in dynamic contractions. The main focus is the interpretation of the signal descriptors in terms of the underlying physiology, whereas technical details on the implementation of the extraction algorithms are omitted. A special issue of the IEEE Engineering in Medicine and Biology Magazine (1) recently has been devoted to advanced algorithms for the processing of EMG during dynamic contractions.
FACTORS THAT INFLUENCE SURFACE ELECTROMYOGRAM IN DYNAMIC CONTRACTIONS
The factors that influence surface EMG has been reviewed previously (8). Most of them are common to both static and dynamic contractions. However, three main features of the dynamic paradigm significantly differ from the static one: (1) the degree of signal nonstationarity, (2) the relative shift of the electrodes with respect to the origin of the action potentials, and (3) the changes in conductivity properties of the tissues separating electrodes and muscle fibers. These three factors are often negligible during static contractions, but can substantially complicate our interpretation of the EMG signal during movement.
A signal is wide sense stationary if its mean value and correlation between samples do not depend on time. In this instance, the power spectrum of the signal also does not vary with time. In static contractions, substantial changes in the statistical and spectral signal properties occur during several seconds. Thus, the EMG signal is assumed to be relatively stationary, and changes can be tracked by sliding an analysis window across the data to analyze the windowed signal as if it were stationary (13). In dynamic contractions, however, signal properties may change at a much faster rate because of rapid recruitment and derecruitment of motor units and changes in joint angle. Classic analysis techniques for stationary signals are often not appropriate during dynamic contractions.
The muscle is not a uniform generator of electric signal. Signals recorded at different locations over the muscle may differ substantially because of heterogeneity in muscle fiber distribution and the generation of the action potentials at the endplates and their extinction at the tendons. Standard electrode positioning is thus necessary in EMG recordings (11). However, when the joint angle changes, the electrodes attached to the skin may shift with respect to the muscle fibers (7) and result in a different relative position at each instant of time during movement. This adds an unwanted signal component that is modulated by the geometrical relation between electrodes and muscle fibers and is thus difficult to predict or remove.
Conductivity of the Tissues
The conductivity of muscle tissue changes when an angular displacement about a joint causes changes in muscle fiber diameter, length, and orientation. Conductivity of muscle depends on the direction of the fibers, with lower conductivity in the direction that is transverse to the fibers. At each point along a muscle, relative fiber direction may change with movement. Thus, action potentials detected at the skin surface can change during movement because of modifications in conductivity of the tissues surrounding the fibers (volume conductor), independent of changes in membrane fiber properties (15).
TECHNIQUES FOR INFORMATION EXTRACTION
Techniques for analysis of dynamic surface EMG can be grouped in two categories according to the type of information extracted: (1) degree of muscle activation (timing of muscle activation and modulation of EMG amplitude) and (2) membrane muscle fiber properties (spectral analysis and conduction velocity estimation).
Timing of Muscle Activation
The simplest application of surface EMG is identifying the timing of muscle activation. This reduces the information content of the EMG to a binary signal indicating the on-off status of a muscle. Many methods are proposed in the signal processing literature to reliably identify the presence of surface EMG (1). Some of them are based on a single threshold applied to the rectified interference signal or to a signal envelope. More advanced approaches consider the evolution of the envelope in time, rather than its instantaneous value, or apply thresholds on the signal after decorrelation of samples.
The requirements on which the detection of muscle activation is based include the following: (1) when the muscle is not active, there is no EMG signal, and (2) when the muscle is active, the EMG signal can be distinguished from noise. Although these requirements can usually be met, there are instances in which it is difficult. The first requirement cannot be met when there is large crosstalk from neighboring muscles (5). This is one of the most important issues in clinical gait analysis. Unfortunately, it is not possible to identify crosstalk with cross-correlation analysis or to remove it with high-pass filtering (8). Double differential recordings, instead of bipolar recordings, can be useful in reducing crosstalk (5), but there is currently no method that can fully eliminate crosstalk. In practice, the presence of crosstalk may be tested in preliminary selective contractions, but this may be subjective and not feasible in all experimental conditions.
The second requirement fails when there is poor signal/noise ratio, especially if the onset and offset of muscle activity are slow. Amplitude cancellation, which occurs because of the summation of positive and negative phases of the action potentials (8), may further reduce the signal/noise ratio and complicate exact detection of the onset and offset of muscle activity. Despite these limitations, the most advanced algorithms currently available allow detection of muscle activity with a resolution of tens of milliseconds, when crosstalk is not present, which is acceptable for most applications (1).
Because absolute EMG amplitude values are not reliable, because of many factors that influence them (8), the signal amplitude is usually normalized, for example, with respect to amplitude during a maximal voluntary static contraction. Analysis of amplitude modulation is performed with the signal envelope (rectification and low-pass filtering) or by estimation of the average rectified or root-mean-square value with a sliding window. Variance of the estimate can be substantially reduced with special techniques, such as signal whitening and multichannel processing (4). However, the main limitation in the interpretation of EMG amplitude is not due to processing algorithms but to the masking effects of unwanted factors.
The use of amplitude modulation for the assessment of relative muscle activation during movement relies on two main requirements: (1) EMG amplitude should be directly related to the level of excitation sent to the muscle from the spinal cord, and (2) amplitude should not be influenced by factors other than the excitation level. Both requirements are difficult to satisfy during dynamic contractions. Amplitude is not directly related to excitation level because of amplitude cancellation (8). Moreover, the relation between amplitude and excitation level depends on the pattern of motor unit activation (10), electrode location in relation to innervation zones and tendon regions, and crosstalk (Fig. 1). In dynamic contractions, volume conductor properties (15) and the relative position of the electrodes with respect to muscle fibers may change over time; thus, amplitude may be additionally influenced by geometrical factors (Fig. 2), in a subject- and muscle-specific way. Quantitative comparisons of patterns of EMG amplitude during movement across muscles or subjects should thus require analysis of the possible confounding factors.
In static contractions, the EMG power spectrum and its modifications over time have been used as indicators of recruitment strategies or fatigue for many decades (13). In dynamic contractions, however, the presence of signal nonstationarity requires the use of more advanced methods. Joint time-frequency representations, including Cohen's class distributions and wavelets, have thus been adopted for the analysis of dynamic tasks (1).
The rationale underlying the use of surface EMG spectral analysis is the scaling effect of action potential conduction velocity on the EMG power spectrum (12). As a consequence, it is often assumed that spectral descriptors, such as mean frequency, are linearly related to the average conduction velocity of the active motor units. This hypothesis is not fully valid even in static contractions (6) and does not hold when the active motor unit pool changes over time. The spectral properties of the surface action potentials are strongly influenced by the distance between the active muscle fibers and the detection point. The action potentials of a newly recruited or derecruited motor unit may contribute to the high- or low-frequency spectral region independent of their conduction velocity. For this reason, trends of spectral descriptors within bursts of dynamic EMG do not correspond to specific recruitment patterns. Similarly, it is not possible to identify the recruitment of different motor unit types from spectral analysis. Moreover, as with estimates of amplitude, spectral features are influenced by modifications in volume conductor properties and the shift of electrodes with changes in joint angle, which result in changes during movement being attributable to both geometrical or physiological effects (Fig. 3) (15). The study of cyclic contractions may minimize this effect if spectral descriptors are estimated over time at the same joint angle and for similar force levels to assess relative, rather than absolute, changes in conduction velocity (myoelectric manifestations of fatigue) (2).
Spectral analysis aims at an indirect estimation of muscle fiber conduction velocity. Because this association is not valid during most movements, a direct measure is necessary. However, this imposes technical difficulties related to the need for multichannel EMG detection during movement. Recently, an approach for recording and processing surface EMG signals to estimate muscle fiber conduction velocity in fast movements has been proposed (9). Direct estimation of conduction velocity in dynamic contractions showed different trends over time with respect to spectral descriptors because of the different effect of the volume conductor on the two estimates (Fig. 4).
Although it avoids some of the limitations of spectral analysis, the direct estimation of conduction velocity during movement is also influenced by methodological factors, such as electrode location. Because electrode location may change over time, conduction velocity values should be compared at the same joint angle, or the electrodes should be placed so that their location is between the innervation zone and tendon for the entire range of analyzed joint angles. This may be accomplished by the assessment of innervation zone and tendon locations at different joint angles (9).
Other Types of Analysis
Many other types of signal analysis have been proposed to extract information from the surface EMG recorded in dynamic contractions, such as recurrent plot or spike analysis (1). Although these techniques are well established from the signal analysis point of view, they lack validation when applied to extract physiologically relevant information from the surface EMG. Simulation studies are needed to better clarify the type of information that can be extracted and the limitations of these promising methods.
Electromyogram Mapping During Movement
Some of the problems in the interpretation of surface EMG during movement may be addressed by placing many electrodes closely spaced over the skin, thus sampling the muscle electrical activity on a larger surface area (high-density surface EMG). Innervation zones and tendons can be identified from the two-dimensional recordings and used as reference points for estimating the shift of the electrodes over the skin with joint angle (Fig. 5). From this information, it may be possible to adaptively select the electrodes used for information extraction to compensate for the change in geometry. The potential of this approach for interpretation of dynamic EMG remains to be explored.
Interpretation of the surface EMG in dynamic tasks is complicated by three main additional factors with respect to static conditions: the signal nonstationarity, the shift of the electrodes relative to muscle fibers, and the changes in the conductivity properties of the tissues separating electrodes and muscle fibers. Detection of the timing of muscle activation is a reliable analysis technique with relevant applications when crosstalk is limited. However, the modulation of signal amplitude is not easily related to the excitation level of the muscle. Spectral analysis provides useful information on the relative changes in muscle fiber conduction velocity (myoelectric manifestations of fatigue) when the pool of active motor units is the same in all intervals of analysis. However, characterization of motor unit recruitment patterns or identification of motor unit type from spectral analysis is not possible. Direct estimation of conduction velocity allows the analysis of motor unit control strategies and fatigue in dynamic tasks, although its application is complicated by technical difficulties in dynamic multichannel recordings. High-density, multichannel EMG is a promising approach for compensating for the geometrical factors that influence signal interpretation in dynamic contractions. In conclusion, although surface EMG recorded in dynamic tasks is a powerful means for assessing muscle function and finds important applications in both clinical and research environments, it has many limitations that remain to be resolved.
I thank Prof. Roberto Merletti (Politecnico di Torino) for the interesting discussions on EMG signal interpretation during my time at the Laboratory for Neuromuscular System Engineering (LISiN), Torino, Italy.
This work was supported by the Danish Technical Research Council (project "Centre for Neuroengineering [CEN]"; contract no. 26-04-0100) and by the European Community (project "Cybernetic Manufacturing Systems [CyberManS]").
1. Bonato, P. (guest editor). Recent advancements in the analysis of dynamic EMG data. IEEE Med. Biol. Mag.
2. Bonato, P., S. H. Roy, M. Knaflitz, and C. J. De Luca. Time-frequency parameters of the surface myoelectric signal for assessing muscle fatigue during cyclic dynamic contractions. IEEE Trans. Biomed. Eng.
3. Campanini, I., A. Merlo, P. Degola, R. Merletti, G. Vezzosi, and D. Farina. Effect of electrode location on EMG signal envelope in leg muscles during gait. J. Electromyogr. Kinesiol.
4. Clancy, E. A., S. Bouchard, and D. Rancourt. Estimation and application of EMG amplitude during dynamic contractions. IEEE Med. Biol. Mag.
5. De Luca, C. J., and R. Merletti. Surface myoelectric signal cross-talk among muscles of the leg. Electroencephalogr. Clin. Neurophysiol.
6. Dimitrova, N. A., and G. V. Dimitrov. Interpretation of EMG changes with fatigue: facts, pitfalls, and fallacies. J. Electromyogr. Kinesiol.
7. Farina, D., R. Merletti, M. Nazzaro, and I. Caruso. Effect of joint angle on EMG variables in leg and thigh muscles. IEEE Eng. Med. Biol. Mag.
8. Farina, D., R. Merletti, and R. M. Enoka. The extraction of neural strategies from the surface EMG. J. Appl. Physiol.
9. Farina, D., M. Pozzo, E. Merlo, A. Bottin, and R. Merletti. Assessment of average muscle fiber conduction velocity from surface EMG signals during fatiguing dynamic contractions. IEEE Trans. Biomed. Eng.
10. Fuglevand, A. J., D. A. Winter, and A. E. Patla. Models of recruitment and rate coding organization in motor-unit pools. J. Neurophysiol.
11. Hermens, H. J., B. Freriks, R. Merletti, D. F. Stegeman, J. Blok, G. Rau, C. Disselhorst-Klug, and G. Hägg. European Recommendations for Surface ElectroMyoGraphy
. SENIAM 8, Roessingh Research and Development, Enschede (NL), 1999.
12. Lindstrom, L., and R. Magnusson. Interpretation of myoelectric power spectra: a model and its applications. Proc. IEEE
13. Merletti, R., M. Knaflitz, and C. J. De Luca. Myoelectric manifestations of fatigue in voluntary and electrically elicited contractions. J. Appl. Physiol.
14. Merlo, E., M. Pozzo, G. Antonutto, P. E. di Prampero, R. Merletti, and D. Farina. Time-frequency analysis and estimation of muscle fiber conduction velocity from surface EMG signals during explosive dynamic contractions. J. Neurosci. Methods
15. Mesin, L., M. Joubert, T. Hanekom, R. Merletti, and D. Farina. A finite element model for describing the effect of muscle shortening on surface EMG. IEEE Trans. Biomed. Eng.