**5. The application of EMG decomposition in the treatment of neuromuscular disorders triggered by stroke**

Stroke, or cerebrovascular accident (CVA), affects a great number of individuals, and is the leading cause of disability among adults [10, 24]. After the event, most individuals must deal with severe reduction of motor functionality [31].

18 Computational Intelligence in Electromyography Analysis: A Perspective on Current Applications and Future Challenges

**Figure 14.** Identification of reference points in time for feature selection. *wo* and *wf* indicate the beginning and end of the region of activity. *p* is the time when the highest peak in the envelope and within the interval [*wo*, *wf* ] occurs, and *po* and *pf* define the boundaries of a rectangular window for

signal is maximal within the RA. This is illustrated in Figure 13.

Computational Intelligence in Electromyography Analysis –

envelope of the RA occurs. This point is also the point where the variation in amplitude of the

*po* and *pf* are respectively the beginning and end of the window to be selected for analysis. These points are defined as follows: *po* = *p* − *min*{*to*, *p* − *wo*} and *pf* = *p* + *min*{*to*, *wf* − *p*}, with *to* set to 2 ms. Note that the width of the window defined by *po* and *pf* may vary for different RAs. Therefore, interpolation (splines) was employed for selection of 40 samples from each window defined in the interval [*po*, *pf* ]. This means that after feature selection each pattern is represented by a 40D vector with samples obtained via an interpolation procedure.

In order to ease the application of the sequence of steps detailed in Figure 10 a graphical user interface (GUI) was devised. The main GUI is shown in Figure 15. The system is capable of importing EMG data organized in columns of a text file and storing them in user-defined variables which are available in a list box. The main interface is organized into four logical

Figure 16 shows the module which allows the user to filter the EMG signal. The filtering procedure based on the Empirical Mode Decomposition is available here. The result of the

The results of the data clustering and visualization step are presented in the GUI shown in Figure 18. At this stage patterns are clustered by means of Generative Topographic Mapping (GTM) and data visualization is performed with the GTM grid [3, 9]. For generation of the GTM grid, a GTM model with 25 Gaussian functions and 16 basis functions with a width of 1 is fitted to the data. The data can also be projected onto the two-dimensional space so that the

Stroke, or cerebrovascular accident (CVA), affects a great number of individuals, and is the leading cause of disability among adults [10, 24]. After the event, most individuals must deal

automatic detection of regions activity is given in the interface shown in Figure 17.

user can visualize the distances of distinct groups of MUAPs (see Figure 19).

**5. The application of EMG decomposition in the treatment of**

feature selection.

278

*4.1.5. Data visualization and clustering*

sections (tabs) that should be accessed sequentially.

**neuromuscular disorders triggered by stroke**

with severe reduction of motor functionality [31].

**Figure 15.** Main graphical user interface of a software that implements the sequence of steps detailed in Figure 10.

**Figure 16.** Graphical user interface which allows the user to filter the input signal.

20 Computational Intelligence in Electromyography Analysis: A Perspective on Current Applications and Future Challenges 280 Computational Intelligence in Electromyography Analysis – A Perspective on Current Applications and Future Challenges EMG Decomposition and Artefact Removal <sup>21</sup>

**Figure 19.** Two-dimensional visualization of groups of MUAPs obtained from the application of

of motor units, muscle contractures, and decreased cortical representation, due to disuse of

EMG Decomposition and Artefact Removal 281

Several rehabilitation therapies have been used in an attempt to recover motor functionality, especially for upper limbs. The majority is based on the premise that the neural system can be retrained [20, 34]. The ability of the neural system to adapt to a new structural and functional condition, as well as its response to a traumatic destructive injury or to subtle changes resulting from the processes of learning and memory, is called 'neuroplasticity"

Recent studies have shown that behavioral experiments, important tools used in rehabilitation strategies based biofeedback techniques, have a strong impact on the motor cortical representation post-stroke. In this sense, we can also infer that it is possible to use biofeedback strategies for the modulation of neural plasticity, seeking the recovery of motor skills in rehabilitation protocols. The question is: how can we generate the appropriate information for feedback in a situation where the encephalic damage is manifested by an uncontrolled recruitment, or the lack of recruitment, of muscle fibers - leading to involuntary muscle hypotonia or hypertonia? Looking for a solution to this problem, a novel strategy, based in the assessment of the recruitment rate of motor units, is under test (see [39]), to generate control information for a multimodal biofeedback system. In this approach, instead of simply evaluating the process of muscle contraction, the researches decided to focus on the neural control over the muscles. However, the information regarding the recruitment of motor units are not readily available through standard surface EMG. Hence, the biofeedback relies heavily

The strategy addressed by [39] is based on the discrimination and feature extraction of EMG signals (Motor Unit Action Potential) in order to control a virtual device. The biofeedback protocol immerses the patient in a virtual reality environment in which a representation of the affected limb will be presented. This virtual member is controlled according to the

Principal Component Analysis.

on a robust EMG decomposition system.

the affected limb [28].

[1, 12].

**Figure 17.** Automatic detection of regions of activity.


**Figure 18.** Data clustering and visualization based on Generative Topographic Mapping [3, 9].

Hemiplegia and spasticity are among the most common post-stroke motor deficits. In general, hemiplegia is characterized by an initial flaccid stage, with motor and sensory losses, in which the patient finds himself unable to sustain or move the affected limb. In many cases, the motor sequel evolves into spasticity, a stage characterized by muscle hypertonia. Hemiplegia and spasticity are closely related to the disuse of the affected limb and to secondary changes in muscles, such as: selective atrophy of fast twitch type II muscle fibers, abnormal recruitment

20 Computational Intelligence in Electromyography Analysis: A Perspective on Current Applications and Future Challenges

**Figure 18.** Data clustering and visualization based on Generative Topographic Mapping [3, 9].

Hemiplegia and spasticity are among the most common post-stroke motor deficits. In general, hemiplegia is characterized by an initial flaccid stage, with motor and sensory losses, in which the patient finds himself unable to sustain or move the affected limb. In many cases, the motor sequel evolves into spasticity, a stage characterized by muscle hypertonia. Hemiplegia and spasticity are closely related to the disuse of the affected limb and to secondary changes in muscles, such as: selective atrophy of fast twitch type II muscle fibers, abnormal recruitment

**Figure 17.** Automatic detection of regions of activity.

Computational Intelligence in Electromyography Analysis –

280

**Figure 19.** Two-dimensional visualization of groups of MUAPs obtained from the application of Principal Component Analysis.

of motor units, muscle contractures, and decreased cortical representation, due to disuse of the affected limb [28].

Several rehabilitation therapies have been used in an attempt to recover motor functionality, especially for upper limbs. The majority is based on the premise that the neural system can be retrained [20, 34]. The ability of the neural system to adapt to a new structural and functional condition, as well as its response to a traumatic destructive injury or to subtle changes resulting from the processes of learning and memory, is called 'neuroplasticity" [1, 12].

Recent studies have shown that behavioral experiments, important tools used in rehabilitation strategies based biofeedback techniques, have a strong impact on the motor cortical representation post-stroke. In this sense, we can also infer that it is possible to use biofeedback strategies for the modulation of neural plasticity, seeking the recovery of motor skills in rehabilitation protocols. The question is: how can we generate the appropriate information for feedback in a situation where the encephalic damage is manifested by an uncontrolled recruitment, or the lack of recruitment, of muscle fibers - leading to involuntary muscle hypotonia or hypertonia? Looking for a solution to this problem, a novel strategy, based in the assessment of the recruitment rate of motor units, is under test (see [39]), to generate control information for a multimodal biofeedback system. In this approach, instead of simply evaluating the process of muscle contraction, the researches decided to focus on the neural control over the muscles. However, the information regarding the recruitment of motor units are not readily available through standard surface EMG. Hence, the biofeedback relies heavily on a robust EMG decomposition system.

The strategy addressed by [39] is based on the discrimination and feature extraction of EMG signals (Motor Unit Action Potential) in order to control a virtual device. The biofeedback protocol immerses the patient in a virtual reality environment in which a representation of the affected limb will be presented. This virtual member is controlled according to the pattern of motor unit recruitment. Thus, although the spastic or hemiplegic patient lacks of proper voluntary control, it is expected that the system will be able to capture small voluntary changes in motor recruitment, as a result of his desire to do so. These changes can then be used to guide the movements of the virtual member. In so doing, the virtual feedback operates as a guide, indicating that the current mental strategy (neuromotor control) is correct and should be encouraged, reinforcing the process of neural reorganization (neuroplasticity).

**Author details**

Slawomir J. Nasuto

Peter J. Kyberd

**7. References**

2006.

8:203–12, 1987.

edition, 1994.

178(9):2176–2193, 2008.

*Uberlândia, Uberlândia, Brazil*

Adriano O. Andrade and Alcimar B. Soares

University of Reading, Reading, UK, 2005.

*Faculty of Electrical Engineering, Laboratory of Biomedical Engineering, Federal University of*

EMG Decomposition and Artefact Removal 283

[1] I Adams. Comparison of synaptic changes in the precentral and postcentral cerebral cortex of aging humans: a quantitative ultrastructural study. *Neurobiology of Aging*,

[2] Metin Akay. *Biomedical signal processing*. Academic Press, Inc, San Diego, California, 1st

[3] Adriano O Andrade. *Decomposition and Analysis of Electromyographic Signals*. PhD thesis,

[4] Adriano O Andrade, Peter J Kyberd, and Slawomir J Nasuto. Time-frequency analysis of surface electromyographic signals via Hilbert spectrum. In Serge H Roy, Paolo Bonato, and Jens Meyer, editors, *XVth ISEK Congress - An Invitation to Innovation*, page 68. [5] Adriano O Andrade, Peter J Kyberd, and Slawomir J Nasuto. The application of the Hilbert spectrum to the analysis of electromyographic signals. *Information Sciences*,

[6] Adriano O Andrade, Peter J Kyberd, and Sean D Taffler. A novel spectral representation of electromyographic signals. In Ron S. Leder, editor, *Engineering in Medicine and Biology*

[8] Adriano O Andrade, Slawomir J Nasuto, and Peter J Kyberd. Extraction of motor unit action potentials from electromyographic signals through generative topographic

[9] Adriano O Andrade, Slawomir J Nasuto, Peter J Kyberd, and Catherine M Sweeney-Reed. Generative topographic mapping applied to clustering and

[11] Alfredo Avellido and Adriano O Andrade. Determination of feature relevance for the grouping of motor unit action potentials through a generative mixture model. *Biomedical*

[12] S N Burke and C A Barnes. Neural plasticity in the ageing brain. *Nature Reviews*

*Society - 25th Annual International Conference*, volume 1, pages 2598–2601. IEEE. [7] Adriano O Andrade, Slawomir J Nasuto, and Peter J Kyberd. EMG signal filtering based on empirical mode decomposition. *Biomedical Signal Processing and Control*, 1(1):44–55,

mapping. *Journal of the Franklin Institute*, 344(3-4 (May-July)):154–179, 2007.

visualization of motor unit action potentials. *Biosystems*, 82(3):273–84, 2005. [10] American Heart Association. Heart disease and stroke statistics — 2011 update: a report

from the american heart association. *Circulation*, 123:e18–e209, 2011.

*Signal Processing and Control*, 2:111–121, 2007.

*Neuroscience*, (7):30–40, 2006.

*School of Systems Engineering, University of Reading, Reading, United Kingdom*

*Institute of Biomedical Engineering, University of New Brunswick, Fredericton, Canada*
