Data sources and searches

This systematic review followed the PRISMA guidelines and the PICO model for the definition of the inclusion criteria: P (Population): “human adults”, I (Intervention): “impact of gait, running or walking on muscle activity of the lower extremities”, C (Control): “healthy or different gait conditions” and O (Outcome): “EMG wavelet parameters” [11]. Neurological diseases were excluded because of the disease-related, specific gait patterns and the possible sensorimotor dysfunction. This systematic review has been listed in the international prospective register of systematic reviews (PROSPERO) with the identification number CRD42016035986.

The electronic databases PubMed/MEDLINE, EMBASE, CINAHL, Cochrane Library, PEDRO, SURF, Art Source, Business Source Premier, Green File and Health Source as well as the search engine Google Scholar were systematically searched with the search terms wavelet AND (EMG OR electromyography) AND (running OR walking OR gait). Additionally, reference lists from retrieved publications were hand-searched. The languages were restricted to English, German, French, Italian, Spanish and Dutch. All study designs, except systematic reviews, were included. The search was completed on June 1^st 2017.

Data extraction and quality assessment

Two investigators experienced in signal analysis performed the whole data extraction process independently. Discussion took place after every fulfilled part of the protocol until consensus was achieved. The investigators independently screened all titles for eligibility, based on the a priori defined PICO inclusion criteria. In case of disagreement, consensus was achieved through discussion. In case of uncertainty, the study was taken to the next step. Afterwards, the same process was performed to screen the abstracts and finally to screen the full texts for eligibility. If no consensus was found after the discussion, a third investigator decided whether the study should be included or excluded. In the further process, the quality of the included studies and the risk of bias were analyzed using “The Cochrane Collaboration`s tool for assessing risk of bias”. The criteria list included six items from which each was scored with “+” if the criteria was fulfilled, with “-”if the criteria was not fulfilled or with “?” if the data was not provided or unclear. According to Viswanathan (2012) [12] the downgrade of the studies not designed as a RCT was limited by rating the selection bias as unclear risk of bias if the method was appropriate.

Data synthesis and analysis

Next, the following data were independently extracted and analyzed by two investigators using a standardized extraction form: First author, year; subject characteristics; intervention; outcome: measures & variables; results; wavelet specification.

Methodological quality

Following “The Cochrane Collaboration`s tool for assessing risk of bias”, all studies showed a high or unclear risk of bias concerning random sequence generation, and blinding of participants and personnel. None of the studies provided information about blinding of outcome assessments. Two studies gave some information about randomization [9,13], 14 studies showed a low risk of incomplete outcome data and 16 a low risk of reporting bias. All studies featured an unclear risk of other bias due to unclear prescription (see Table 1). Moreover, most of the studies had a small sample size (N<30).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Cochrane risk of bias analysis.

https://doi.org/10.1371/journal.pone.0206549.t001

Study characteristics

In 18 studies, EMG activity patterns of the lower extremities were investigated: 13 investigated walking, 4 running and one different walking and running conditions. The summary of wavelet analysis of EMG walking data is depicted in Table 2 and of EMG running and different conditions data in Table 3. In the included studies, three main topics were discussed: The recognition rate, time period characteristics, and motor unit characteristics. These three topics are defined in the following section and summarized in Tables 2 & 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Data extraction: Summary of wavelet analysis of EMG of lower extremity muscles while walking.

https://doi.org/10.1371/journal.pone.0206549.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Data extraction: Summary of wavelet analysis of EMG of lower extremity muscles while running and different conditions.

https://doi.org/10.1371/journal.pone.0206549.t003

Recognition rate

The recognition rate is the capability of a method to correctly assign participants to a specific group [14,15]. Different classification procedures were applied to find the boundaries between two groups under investigation and to improve the accuracy of the method. Intensity patterns which reflect how strong a signal is at time t, are sometimes very difficult to assign to a certain condition by visual inspection because the signals can appear inconsistent [1,16]. However, differences between the measured conditions must be reflected in the signal. The combination of wavelet analysis with methods from multivariate statistics and the field of pattern recognition, like principal component analysis (PCA) and support vector machines (SVM) [16], allows to detect subtle differences in muscle activation patterns. PCA is a statistical method to identify components and order them according to their strength to explain the variance in given datasets. It allows reducing the dimensionality in large datasets and hence it also allows uncovering the most important factors. The term “support vector machine” is inspired from their application in machine learning applications, for example face or speech detection by computers. SVMs are suitable for allocating a given set of data to one of two classes, for example to a patient or a control group. Wavelet analysis in conjunction with subsequent PCA to find principal patterns in EMG intensities, was used to discriminate between healthy people and patients with ankle osteoarthritis while walking [14], between barefoot runners and shod runners [16] and males from females [9]. The application of wavelet analysis together with SVMs was used to find the boundaries to discriminate patients with total knee arthroplasty from healthy people while walking [15] or to correctly assign the EMG patterns of different running tasks to each subject [17,18]. Eight studies computed the recognition rate [9,14–19].

Recognition rate distinguishing between different conditions

The recognition rate discriminated between different running, walking or footwear conditions measured with different muscles of the leg [9,15,17–19]. The maximum recognition rate was 78 to 94.4% and varied between different muscles and methods. Jaitner et al. (2010) [18] tested the recognition of individual EMG patterns of leg muscles with different walking and running conditions with a resulting rate of 92.9 to 100%. Moreover, Kuntze et al. (2015) [15] discriminated total knee arthroplasty patients from healthy people with a rate of 68.4–73.7% while walking. Von Tscharner (2003) [9] determined more than 95% correct gender classification as male or female runners. The recognition method was able to extract gender specific differences in timing, intensity and frequency distribution of electromyographic signals.

Time period characteristics

The behavior of EMG intensity and frequency over time was analyzed. The non-stationary wavelet approach was used for the evaluation of muscle timing, time shifts and early or delayed muscle activation [6,10,13,15,16,20–23]. Originally, wavelet analyses have used a mother wavelet which waslinearly scaled. However, this method did not represent the physiological response time of the muscles in the higher frequencies [1]. In contrast, the time resolution of non-linearly scaled wavelets described by von Tscharner et al. (2000) [1] represent a more physiological condition. Differences were revealed between the activation patterns of walking versus running.

Time period characteristics of the lower extremity muscles while walking

In healthy subjects walking presented a pre-activation peak 50–100 ms before heel-strike [10], followed with a sharp decline as a response to the reflex attenuation. This pre-activation was highly subject-specific. For the reflex response with a reaction time to maximum activity between 30–50 ms after heel-strike, similar muscle activation patterns were found. Thus, the reflex response was less subject-specific than the pre-activation [10]. Patients with total knee arthroplasty presented with a delayed peak activity of vastus medialis and biceps femoris muscle in the early stance phase of walking but with a prolonged activation in mid stance of vastus medialis, vastus lateralis, and biceps femoris [15]. The control strategy of a joint while walking, requires a dynamic interplay between muscles, the central nervous system and external factors [20]. According to Huber et al. (2011) [20], neuromuscular control mechanisms showed a precise pacing to control the individual walking pattern. This pacing of neuromuscular activity while walking, seemed to be controlled as a programmed function at 40 ms intervals, independent from subject and muscle. Nonetheless, the neuromuscular rhythm was less precisely controlled while walking than while running, reflected as many variations in timing and intensity of muscle activation [20].

Time period characteristics of the lower extremity muscles while running

Electromyographic signals have been found to be very subject-specific [16,23]. Therefore, EMGs of different muscles have been arranged into Multi-Muscle activation Patterns (MMPs). To form a MMP the intensity patterns of individual muscles have been stacked one above the other. Thus, the MMPs represented the intensity of the electromyographic signals of several muscles simultaneously. These resulting MMPs demonstrated a pattern with high similarity of muscle activation while running [6]. The EMG activation of the gastrocnemius presented between 0 (heel-strike) and 200 ms after heel-strike and of the tibialis anterior between -100 and 150 ms [13]. Altered shoe conditions influenced the timing of tibialis anterior minimally in pre-heel-strike anticipation but delayed for the post-heel-strike reflex. There was a shift for the maximum EMG intensity from 25 ms after heel-strike in the barefoot condition to 60 ms in the shoed conditions [21]. Between different running speeds differences have been presented in motor unit recruitment patterns. Overall, the higher the running speed the higher the intensity for all muscles tested [22].

Motor unit characteristics

Recruitment patterns of slow type I to fast type II fibers have been investigated by means of the intensity levels at different wavelet frequency bands [13]. Intensities below 100Hz simultaneously reflected the activity of slow and fast muscle fibers; above 100Hz fast muscle fibers dominated [13].

Motor unit characteristics of the lower extremity muscles while walking

Down shifted EMG spectra of tibialis anterior toward lower frequencies have been observed in patients with ankle osteoarthritis. The wavelet patterns of patients showed a trend towards less intensity production of tibialis anterior between 170 and 271Hz but more intensity between 19 and 62Hz [24]. Atrophic muscles have been demonstrated to present relatively less intensity in high frequency bands and more in the lower frequencies [14,24,25]. Textured insoles caused a reduction in the intensity of the electromyographic signal of the soleus and tibialis anterior in the low (38-62Hz) but not in the high frequency (170-271Hz) domains and this continued throughout the entire stance phase [26].

Motor unit characteristics of the lower extremity muscles while running

Motor unit characteristics while running have been mainly investigated by Wakeling et al. and von Tscharner et al. [6,13,21–23]. They showed that the tibialis anterior was activated by the fast fibers (60-270Hz) in the pre-heel-strike period followed by a drop in intensity at heel-strike. The slow fibers (10-90Hz) controlled the early part of the stance phase. The two spectrally very different time periods were only 15 ms apart [13,21]. Wearing shoes resulted in an increase of the pre-heel-strike activity of the tibialis anterior and in a decrease of post-heel-strike activity of its slow muscle fibers [26]. Changes in shoe hardness resulted in changes for the tibialis anterior in the intensity of the EMG, both positively and negatively at high (192-301Hz) and low frequency (11-75Hz) bands in a subject-specific manner. These changes were consistent with changes in the recruitment patterns between fast and slow motor units [23]. The activation of the gastrocnemius started with low frequency components after heel-strike and shifted gradually to higher frequency components. The highest frequencies were recorded 110 ms after heel-strike. However, 160 ms after heel-strike, predominant activities of slow muscle fibers were recorded [13]. The mean frequency of hamstrings dropped from 60Hz at the pre-heel-strike maximum to 40Hz at heel-strike. At 120 ms, a slightly higher maximum was reached compared to the pre-heel-strike maximum. The mean frequency of the quadriceps femoris had already increased during pre-heel-strike [9]. During 30 minute running trials, the low frequency components decreased and the high frequency components increased in intensity [6,22]. Regarding gender differences, women activated more high frequency components of tibialis anterior, between -300 and -70 ms, but less between -70 ms and heel-strike than men. Women tended to activate muscle fibers in the lower frequency components [9].

Risk of bias

Since none of the studies reviewed were designed as randomized control trials, the issues of randomization and blinding are not relevant. The study designs of the included studies were appropriate regarding the research questions.

Recognition rate

Non-linearly scaled wavelets, as originally described by von Tscharner (2000) [1] have been shown to discriminate not only between walking and running, but can also detect difference in muscle activity during ambulation in patients with different orthopedic conditions [9,14–19]. Therefore, these techniques might be clinically useful to detect locomotor dysfunction, as well as help design therapeutic interventions (such as orthotic devices, insoles, etc.) to treat or prevent musculoskeletal problems of the lower extremities.

Time period

The use of wavelets for processing non-stationary EMG has been shown to be suitable for evaluating muscle timing, time shifts and early or delayed muscle activation [6,10,13,15,16,20–23]. The major sources of the variability in walking [10,14,15,20] and running [6,13,16,22,23] could be evaluated with wavelets. It was shown that the heel-strike event is an important source of this variability as each walking or running step requires a special adaption to the given condition [10]. However, the neuro-muscular control system reacts at each heel-strike with muscular preparation (tuning) and adaptation. Wavelet analyses made it possible to analyze this muscular preparation a few milliseconds before heel-strike and the muscular adaption after heel-strike considering time, intensity and frequency content. The neuromuscular control mechanisms showed a precise pacing at 40 ms intervals to control the individual walking pattern [20]. This periodic, cortical rhythm (Piper rhythm) seemed to represent the fine tuning needed for controlled movements [20]. Maurer et al. (2013) found that the Piper rhythm varied with running speed [27]. It was supposed that the frequency adapted to exterior condition and modulated the muscle activity between on and offset. The mean frequency of the Piper rhythm increased from 28 to 36 Hz as running speed decreased from 4.9 to 1.3 m/s. The authors assumed that very slow running speeds are not natural and therefore, require more efficient control [27]. However, the neuromuscular rhythm was less precisely controlled while walking than while running, reflected as many variations in timing and intensity of muscle activation [20,27]. These findings are consistent with the assumption that running is more tightly controlled and less variable [27].

This knowledge of muscular preparation and adaption to variable heel-strike events, as well as the interplay of the muscles are useful regarding training protocols in sports and in rehabilitation [25]. It allows athletes and therapists to monitor the training of specific muscles, to optimize the movements and to observe the training progress [25]. Muscles of patients with ankle osteoarthritis or total knee arthroplasty were activated earlier and for longer periods than in healthy people [14,15]. It was assumed that the central nervous system would have to activate the selected motor units of these patients earlier to obtain the required force at heel-strike because of the lower conduction velocity or the longer electromechanical delay of the selected fibers [14,28]. Furthermore, as significant differences were found in the lower extremity muscles [22], the influence of different locomotion speeds and gait variations should be considered in further studies.

Motor unit characteristics

Fast type II muscle fibers generate higher rates of force development than slow type I fibers, and are more likely to allow a fast release compared to slow fibers [21]. During the impact of heel-strike, a fast development of force in the tibialis anterior muscle is needed to stabilize the joints. After heel-strike a fast relaxation is important as well to release the forefoot. This explains that the highest frequencies have been found just before heel-strike [21]. However, task specific recruitment of the slower, fatigue resistant muscle fibers have been preferentially recruited while walking [26]. It is supposed that fast and slow twitched fibers are recruited to match their contractile properties to the mechanical requirements for motion, and these requirements change between gait mode, speeds and within stride [22]. Changes in shoe materials resulted in changes in the intensity ratio of the EMG at high and low frequency bands [23]. Wakeling et al. (2002) interpreted the findings as a response to altered impact forces. The spectral changes of the tibialis anterior muscle towards lower frequencies in patients with ankle osteoarthritis, has been associated with a selective atrophy of type II muscle fibers. Changes in muscle fiber recruitment while walking or running may provide information for prevention and treatment of runners. However, to verify this assumption, histological studies (not only in animals as used as theoretical basis in the studies) are needed [14,24]. A strong and critical debate has been held on this topic. Farina (2008) criticized that von Tscharner et al. did not discuss and consider the electrophysiological properties associated with different fiber types [29]. Furthermore, it was mentioned that the EMG power spectrum can be influenced by many different factors and therefore, the influence of selected physiological variables will probably be small. Farina (2008) noted that the use of spectral analysis of EMG to measure motor unit recruitment or the properties of muscle fiber types has not been validated [29].

Strengths and limitations

To the best of our knowledge, this is the first systematic review that summarizes the use of wavelet analyses of electromyographic recordings from lower extremity muscles during walking, running, and certain orthopedic conditions. The influence of the interdisciplinary knowledge of the team including physiotherapists, sports and movement scientists, biomechanics and engineers, all experienced in signal analysis is another strength of the review.

However, the study has the limitation that in a couple of studies low numbers of participants were included. Furthermore, almost all reviewed articles applied the non-linearly scaled wavelet method developed by von Tscharner (2000) [1]. However, other kinds of wavelets with other time/frequency resolution might resolve other aspects of the interplay of muscles with different dynamic motor tasks. This aspect should be considered in further studies.