The aim of the study is to investigate the capacity of post-stroke individual (study group) to modulate their EMG muscle activation pattern (MAP) compared to healthy individuals (control group), and to correlate these capacities with their motor impairments. Twenty post-stroke individuals and 12 healthy individuals will participate in this study. each participant will carry out hand-reaching movements in multiple directions, monitored by an EMG device. The modulation of the EMG signal will be compared between groups in terms of EMG-MAP and in terms of muscle-synergies. Additionally the MAPs and synergies of the study group will be correlated with their Fugl-Meyer (FM) assessment scores. Analysis of the muscle synergies underlying the EMG signal will be carried out by the Non-negative Matrix Factorization (NMF) algorithm.

Introduction: Upper extremity function after a stroke is impaired and characterized by abnormal, stereotypical and uncoordinated movement pattern. Decreased neural drive in the damaged corticospinal system causing a decreased agonists motor units firing, spasticity, impaired motor coordination. A more comprehensive understanding of the way our brain controls and regulates limb-movements, through the spinal cord, may enhance more advanced rehabilitation techniques.

Current concept in motor control suggest that the brain-cortex modulates and synchronizes the activation of discrete number of functional units within the brainstem and spinal cord. These neural functional units i.e. muscle synergies, when linearly combined facilitate the production of diverse limb-movements. This control mechanism may in a large extent explain the way the CNS reduces the dimensionality of the vast number of degrees of freedom embedded in the CNS to a discrete number of muscle synergies. Therefore, execution of a movement may only requires to linearly combine these synergies and regulates its intensity of activation along the time domain.

The existence of such a control mechanism attracted the attention of both clinicians and scientists to use its properties to enhance motor recovery after a stroke. Therefore, studies emerged to investigate how cortical damage impact the synchronization of synergies, and also whether it changes the internal structure of the synergies. Despite numerous studies in this domain, there is lack of consensus regarding how stroke impact this control mechanism, and extent of correlation between the level of impairment and the synergy structure. The study objectives are to compare the synergy structure and the MAP in hand-reaching movements in multiple directions between post-stroke individuals and healthy individual, and to correlate between these properties and the motor impairments in post-stroke individuals.

Methods:

Participants: Twelve healthy volunteers (control group) and 20 post-stroke individuals (study group) will participate in the study. Inclusion criteria for the study group will be individuals, above the age of 20, who sustained unilateral cerebral stroke, with hemiparesis. Exclusion criteria for the trial are sensory aphasia, unilateral neglect and presence of other neurological disease such as Parkinson's disease or Alzheimer's disease.

Equipment: The Hand-reaching Spatial Device (HRSD) is an adjustable, simple tool allowing standardization of hand pointing movement for 9 different directions between different participants. It is composed of two vertical rods to which are attached three semi-circular shelves. Each shelf contains three movable pointing pins that can be adjusted left and rightward to accommodate the variable arm length of each participant. The lowest shelf was located 10 cm above the table, the middle was located 35 cm above the table and highest 55 cm above the table. For each participant the HRSD was located at the maximum hand reach distance in front of the tested shoulder. The side pins were located at a 45-degree angle to the shoulder joint to both sides. The arrangement of the targets on the HRSD was designed to cover the majority of hand-reaching movements.

Surface EMGs were recorded (Trigno 8, Delsys, Boston, MA) from 8 muscles of the shoulder girdle and arm: trapezius (TRS); deltoid anterior (AD), medial (MD), and posterior fibers (PD); and pectoralis major (PECT); infraspinatus (IS); biceps (BI); triceps (TRI). Electrodes were placed in accordance with the guidelines of the Surface Electromyography for the Non-Invasive Assessment of Muscles-European Community Project (SENIAM) [34]. Maximum voluntary contractions (MVCs) were performed prior to data collection to verify correct electrode placement and for normalization. One-minute rest periods followed each MVC to limit the possibility of fatigue. EMG signals were band-pass filtered (20-450 Hz), and sampled at 2000 Hz.

Protocol: The MVC was measured by standard muscle testing [35]. Then the participant sat in front of a table with his forearm resting in a comfortable position. The HRSD was located as mentioned above. Participants were requested to point to each target 5 times according to voice prompting that was activated by the EMG software every 10 seconds, for 45 pointing movements. The order of pointing targets was constant for all the participants.

Data analysis EMG preprocessing Data analysis was performed using Matlab (The MathWorks, Inc.). EMGs were demeaned, follow by RMS calculation using overlapping window of 50 samples (25 millisecond around each timepoint). Mean baseline EMGs for each trial were subtracted from the averaged data for the sequence of reaching movements. Hence, the EMG data for each trial, a vector whose dimension was 8 (the number of muscles recorded), corresponded to active force generation beyond any residual baseline muscle activity. The EMG data was normalized in accordance to the Maximal Isometric Contraction (MVC) for each muscle.

The NMF algorithm originally used by Lee and Seung (1999 and 2001), was applied to identify muscle synergies and their activation weights. An EMG pattern recorded in hand-reaching movements was modeled as a linear combination of a set of N muscle synergies, each of which specified the relative level of activation across 8 muscles, and activated by a time-varying activation coefficient:

V^(M×T)≈W^(M×N)∙H^(N×T) (4) Where V is the EMG data set matrix with M as the number of muscles (8 muscles), T as the number of time samples, W is the synergy matrix and H is the coefficient matrix. W is m×n is a matrix with n synergies, m is the number of muscles, and H is the n×t matrix of synergy activation coefficients. Thus, each column of W represents the weights of each muscle for a single synergy, and each row of H represents how much the corresponding synergy was activated or used to generate force. In this model, it is possible for each muscle to belong to more than one synergy and thus the EMG of any single muscle might be attributed to simultaneous or sequential activations of several muscle synergies.

In order to determine the optimal number of synergies for the whole group, the EMG data of all the targets were concatenated for each participant. Then the EMG's of the whole sample were concatenated before applying the NMF. The optimal number of synergies (d) was defined as the number of synergies that captured the highest of the total variance of the data, suggesting that additional synergies only captured small residual amounts of variation attributable to noise. This procedure allowed us to estimate the optimal number of synergies for the whole sample to execute any reaching movement in space regardless of the direction of the movement.

The NMF algorithm required the number of synergies extracted to be specified before the application of the algorithm. Therefore, for each data set, the VAF was calculated while changing the number of synergies from 1 to 7. The VAF was calculated using the equation:

VAF(H)=100%×(1-(|(|V-WH|)|_2^2)/(|(|V|)|_2^2 )) (6) Where V is the original matrix, and W and H are the derived, factorized matrices.

Generalization of movement directions

The aim at this stage of analysis was to establish whether a set of discrete number of synergies exist that control any reaching movement in space. Therefore, it was investigated how movement in certain directions could account for movements in other directions. The EMG data for each movement direction was pooled separately across the 8 muscles and concatenated it for the whole sample. In that way the derived set of synergies would have to account for the variance between different subjects, but would also be specific for that direction alone. The NMF was applied separately for each movement direction according to the equation:

V_i≈W_i∙H_i (7) where i is the target number, which corresponded to specific movement direction in space. In this stage of analysis V_i (the EMG matrix) was given as an input for each target, i∈[1,9], and matrices W_i,H_i were updated iteratively. The study procedure included reaching for 9 different target directions in space, allowing us to further investigate if there was a single set of synergies that could account for movements in other directions.

This was done by using a cross-validation technique between the V_i matrices and the W_j matrices by applying a modified version of the NMF algorithm, followed by corresponding VAF calculation changing the number of synergies (d) from only 3 to 5, and not from 1 to 7 based on the results of the NMF for all the participants and for all targets, as detailed in the results section. In the modified version of the algorithm, both V_i and W_j (the synergies matrix) were given as an input. Only the H_(i,j) coefficients matrix of target i, was updated and outputted.

The cross validation process of the modified NMF was carried out for each combination of a data matrix V_i (of target i) and a synergy matrix W_j (of target j), resulting in 9×9 matrices H_ji. For every i,j∈[1,9], we factorize V_i such that W_j H_ji≈V_i.

The reference set of muscle synergies was chosen by calculating the VAF for each of the 9×9 factorizations:

VAF(H_ij )=100%×(1-(|(|V_i-W_j H_ij |)|_2^2)/(|(|V_i |)|_2^2 )) (8) assuming that consistent high values of VAF(H_ij) for a specific V_i may indicate that the synergies obtained from movements in this direction may accurately explain movement in other directions.

Thus, for each predefined number of synergies (d) a 9×9 matrix was received in which each cell represented the accountability of a given synergy (row) to a specified direction (column). Each row in the resulting matrix represented the overall "performance" of the appropriate set of synergies, and so the row with the highest average VAF was chosen for the next stage of analysis.

Direction modulation of muscle synergies Once the set of synergies (W_j ) was chosen, setting the activation coefficients for every target (H_ij,when i∈[9,1]), it was determined which synergies are dominant for each of the directions. For each number of synergies, the mean activation coefficient of every synergy for every direction was calculated. Setting the number of synergies to 4, for example, resulted in 9 vectors (one for each movement direction) of 4 values, representing the 4 synergies. Then, the average amplitude of each of the synergies across the direction of movement, and between movements in different directions across synergies, were measured.

Inclusion Criteria:

• Cortical stroke (study group)
• hemiparesis

Exclusion Criteria:

• sensory aphasia
• complete arm paralysis
• unilateral neglect
• other neurological disease (PD, Alzheimer)