Variation of Muscle Coactivation Patterns in Chronic Stroke During Robot-Assisted Elbow Training

Methods 

After obtaining approval from the Human Subjects Ethics Sub-Committee of the Hong Kong Polytechnic University, we recruited 7 hemiplegic subjects after stroke for the study. All of the subjects were in the chronic stage (at least 1y postonset of stroke; 6 men, 1 woman; age, 51.1±9.7y). All subjects received a robot-assisted elbow training program consisting of 20 sessions, with at least 3 sessions a week and at most 5 sessions a week, and finished in 7 consecutive weeks. Each training session was completed in 1.5 hours. Before and after the training, we adopted 2 clinical scales to evaluate the voluntary motor function of the paretic upper limb (the elbow and shoulder) of the subjects: the Fugl-Meyer Assessment (FMA; for elbow and shoulder; maximum score, 42)18 and the Motor Status Scale (MSS; shoulder/elbow; maximum score, 40).19 Spasticity of the paretic elbow of each subject before and after the training was assessed by the Modified Ashworth Scale (MAS) score.20 The clinical assessments of this study were conducted by a blinded therapist.

During each training session, each subject was comfortably seated, and the affected upper limb was placed horizontally on an electromyography-driven motor system (a Dynaserv motor,a associated with an AKC-205A torque sensor; accuracy, .03Nmb) developed by Song et al21 with the elbow joint positioned at the origin, as shown in figure 1. The forearm of the affected side was placed on a manipulandum, which could rotate with the motor; and the elbow angle signals were measured by the motor via readings of the positions of the manipulandum. A belt was used to fasten the shoulder joint in order to keep the joint position still during elbow extension and flexion. Electromyography electrode pairsc with a center separation of 2cm were attached to the skin surface of the muscle belly of biceps brachii, triceps brachii (lateral head), anterior deltoid, and posterior deltoid, according to the configuration specified in Cram’s work.22 The electromyography electrode pairs were not moved once placed. The electromyographic signals were preamplified,d band-pass filtered (from 10 to 500Hz) and recorded through an analog-to-digital card,e together with the angle signals, with a sampling frequency of 1000Hz.

View Large Image Download to PowerPoint

Fig 1. The experimental setup used for elbow extension and flexion training. Subjects conducted elbow flexion and extension on the horizontal plane, by tracking a target cursor moving at 10° per second on the screen. Electromyography (EMG) electrode pairs were attached to the skin surface of the muscles biceps brachii (BIC), triceps brachii (TRI), anterior deltoid (AD), and posterior deltoid (PD). After preamplification and filtering, the electromyographic signals were digitized and stored along with the angle and torque signals. The electromyographic activity of triceps brachii was also used as a control signal to the electromyography-driven motor system for providing the supportive torque during the training. Abbreviation: A/D, analog to digital.

The electromyographic signals for the muscles of interest during the resting state were first recorded before any voluntary motion taken by a subject in each session, which served as the electromyographic baselines of the individual muscles for the session. The isometric maximum voluntary flexion (IMVF; duration, 5s) and extension (IMVE; duration, 5s) of the elbow at a 90° elbow angle were then measured at a repetition of 3 times, respectively, with a 5-minute rest break after each contraction to avoid muscle fatigue. Other studies have reported that the maximum extension and flexion torques could usually be found at 90° of the elbow angle for hemiplegic subjects after stroke.23 During the training, each subject was required to carry out voluntary elbow flexion and extension in the elbow range from 0° to 90° (0° representing full extension) by tracking a target cursor moving at an angular velocity of 10° per second on the screen for both flexion and extension. From the study by Cheng et al24 and our own experimental experiences,21 10° per second was chosen as a reasonable speed for subjects after stroke to follow, in order to prevent too difficult or too easy a pace for the subjects to achieve. Background music and verbal encouragement were given to subjects during the training. During tracking, active-assisted torques were generated by the motor system during the extension only,21 based on the robot design, with the supportive torque controlled by electromyographic signals.24 The active-assisted torque during the extension movement was defined as:

(1)

(2)

Electromyographic activity from the muscles of interest and angle signals during the training were recorded and stored in a computer during the even-numbered sessions of the training for offline processing. The elbow angle signals were low-pass filtered with a cutoff frequency of 20Hz. The torque signals during the IMVF and IMVE were also low-pass filtered with a cutoff frequency of 10Hz. A 4th-order, zero-phase forward and reverse Butterworth digital filter was adopted for the filtering processes. Figure 2A shows the representative signals recorded from a subject during the training. The linear envelope of the recorded electromyographic signals was obtained by (1) full-wave rectification, (2) low-pass filtering (10Hz cutoff frequency with fourth-order, zero-phase forward and reverse Butterworth filter), (3) subtraction of the average electromyographic activity during the resting state, and (4) normalized to the maximum value of electromyographic activation in each muscle during either a training session or the IMVF and IMVE of each session.27 Most of the electromyographic maximum values were observed in IMVF and IMVE, and only few (4 sessions) were found during the tracking task. The coactivations among muscle pairs during the training were studied by the cocontraction index (CI) as introduced in Frost et al’s study,27 that is,

(3)

View Large Image Download to PowerPoint

Fig 2. Representative signals captured in the training experiment. (A) The raw electromyographic signals from the triceps brachii and biceps brachii, and torque signals during IMVF and IMVE after the filtering; and the biceps brachii and triceps brachii electromyographic signals and the angle signals during a tracking trial. (B) Illustration of the calculation of cocontraction indexes (CI) (with the unit of %Max) of the muscle pairs during a tracking trial.

Results 

Table 1 shows the clinical scales used for the impairment evaluation before and after the 20-session training. It was found that the mean values of scores for FMA and MSS had increased significantly (t tests, P<.05) and the mean value of MAS decreased significantly (t tests, P<.05) after the training. For each subject, the clinical scales indicated improvements in at least 2 items.

Table 1. The Clinical Scales Measured Before and After the Training

		Pretraining Scores			Post-Training Scores
	Subject	MAS	FMA	MSS	MAS	FMA	MSS
	1	1	15	15	1	17	19.6
	2	2	14	16.4	0	17	18.6
	3	1+	14	14	1	14	17.6
	4	2	14	12	1+	15	16
	5	2	12	18.6	1	15	18.6
	6	2	14	12.6	1	14	14.4
	7	2	10	14.4	1+	14	21
	Mean ± SD	1.8±0.4	13.3±1.7	14.7±2.3	0.9±0.5	15.1±1.3	18.0±2.2

Abbreviation: SD, standard deviation.

Figure 3 shows the variation of the overall RMSE of the elbow angle during the tracking training. The overall RMSE varied significantly across the sessions with a decreasing tendency (1-way ANOVA, P<.05). The statistically significant decrease occurred at the 10th session compared with the values for the 2nd, 4th, and 6th sessions (post hoc tests, P<.05). The variation of RMSE from the 2nd to 8th sessions was not significant. There was also no significant changes in RMSE from the 10th to 20th sessions, except a local minimum at the 19th session (post hoc tests, P<.05). Decreasing tendencies in mean RMSE value were also observed in all individual subjects by comparing the mean RMSE values of the 2nd and 20th sessions and the decreases varied from 15.6% (subject 6) to 59% (subject 3). For subjects 1, 2, 3, 4, and 7, the maximum RMSEs were observed at the 2nd session, whereas for subjects 5 and 6, the maximum RMSEs appeared at the 6th session.

View Large Image Download to PowerPoint

Fig 3. The variation of the mean RMSE values of individual subjects (from S1 to S7, represented by symbols shown in the legend); and the global RMSE summarizing the performance of all subjects during the tracking across the training sessions, represented by the values of mean and standard deviation (SD) (error bars).

Figure 4 shows the electromyographic activation levels of each muscle during the training. The overall electromyographic activation level of the 4 muscles varied significantly across the sessions during the training (1-way ANOVA, P<.05). A significant decreasing tendency in the overall electromyographic activation level for the biceps brachii, triceps brachii, and anterior deltoid were found by comparing the maximum value (observed at the 4th session for the biceps brachii, at the 8th session for the triceps brachii and anterior deltoid) with the value at the last session (post hoc tests, P<.05). Decreases in the mean electromyographic activation level of the biceps brachii, triceps brachii, and anterior deltoid for the individual subjects were also found, varying from 3.3% (subject 2, triceps brachii) to 84.7% (subject 7, biceps brachii), with the maximum values appearing in or before the 10th session. The result showed the significant decreases of the overall electromyographic activation level for the biceps brachii and triceps brachii mainly occurred before the 12th session (post hoc tests, P<.05). There was no significant decrease in the overall electromyographic activation level of the biceps brachii from the 10th session till the end, and the overall activation level of the triceps brachii did not change from the 12th to 20th sessions. The decreasing tendency of the overall anterior deltoid activation level was significant from the 8th session to the 20th session (post hoc tests, P<.05). There was no decrease found in the overall posterior deltoid activation level throughout the training; and the electromyographic levels at the 6th, 8th, and 20th sessions were even significantly higher than that of the 2nd session (post hoc tests, P<.05). The posterior deltoid activation levels for individual subjects did not display a consistent trend.

View Large Image Download to PowerPoint

Fig 4. The variations of electromyographic activation level for the biceps brachii, triceps brachii, anterior deltoid, and posterior deltoid muscles. The electromyographic activation levels from individual subjects (ie, S1 to S7) are represented by their respective symbols shown in the legend. The global electromyographic activation levels of the muscles, summarizing the performance of all subjects, are represented by values of mean and SD (error bar).

Figure 5 shows the muscle cocontraction patterns during the training, represented by the cocontraction index of each muscle pair. The variations in the overall cocontraction index of all muscle pairs were significant (1-way ANOVA, P<.05), and the overall cocontraction index of all muscle pairs reached their maximum at the 8th session. There was no significant change found in the overall cocontraction indexes of the triceps brachii and posterior deltoid and biceps brachii and triceps brachii (see fig 5) during the first 8 sessions. The overall cocontraction indexes of the biceps brachii and anterior deltoid, anterior deltoid and posterior deltoid, and biceps brachii and posterior deltoid at the 8th session were significantly higher than the cocontraction indexes at the 2nd session (post hoc tests, P<.05). The overall cocontraction indexes of the muscle pairs biceps brachii and anterior deltoid, anterior deltoid and posterior deltoid, and triceps brachii and anterior deltoid reached a local minimum (post hoc tests, P<.05) at the 6th session before the appearance of the maximum mean values at the 8th session. For all muscle pairs, there was a significant decrease in the cocontraction index value from the 8th session to the 10th session (post hoc tests, P<.05). After the 8th session (from the 10th to 20th sessions), the overall cocontraction index values of the biceps brachii and triceps brachii, biceps brachii and anterior deltoid, anterior deltoid and posterior deltoid, and triceps brachii and anterior deltoid showed a significant decreasing tendency until the end of the training (post hoc tests, P<.05). The overall cocontraction indexes of the triceps brachii and posterior deltoid and biceps brachii and posterior deltoid varied nonsignificantly from the 10th session to the 20th session. The maximum values of the cocontraction indexes of the muscle pairs for each subject appeared mostly on or before the 8th session, except the cocontraction indexes of biceps brachii and anterior deltoid for subject 2 (at the 12th session), anterior deltoid and posterior deltoid for subject 2 (at the 10th session), triceps brachii and anterior deltoid for subject 4 (at the 16th session), and biceps brachii and posterior deltoid for subject 6 (at the 10th session). By comparing the maximum cocontraction index value with the cocontraction index at the last session, decreases in the cocontraction indexes of the muscle pairs for the individual subjects were found to vary from 7.6% (biceps brachii and posterior deltoid for subject 1) to 82.5% (biceps brachii and triceps brachii for subject 7).

View Large Image Download to PowerPoint

Fig 5. The variations in cocontraction index of the muscle pairs, triceps brachii and posterior deltoid, biceps brachii and triceps brachii, biceps brachii and anterior deltoid, anterior deltoid and posterior deltoid, triceps brachii and anterior deltoid, and biceps brachii and posterior deltoid. The cocontraction indexes of individual subjects are represented by symbols shown in the legend. The values of the global cocontraction indexes for a muscle pair summarizing the performance of all subjects are represented by the values of mean and SD (error bar).

Discussion 

After the 20 sessions of robot training of the elbow extension/flexion, motor improvements could be observed in all subjects, associated with the improved clinical scores, and decreases in the RMSE, cocontraction indexes, and electromyographic activation levels. The decrease in the clinical score of MAS suggested a reduction in spasticity of the impaired upper limb of the subjects. Increases in the FMA score and MSS implied improved motor functions of the paretic upper limb during prescribed voluntary movements. However, pre- and post-tests using the clinical scales can only provide general observations on the functional improvements, with little information revealed about the recovery process associated with the specific treatment.

The decrease in the mean values of the RMSE during tracking across the sessions in all subjects (see fig 3) suggested improvement in tracking performance. Adult cerebral cortex is capable of significant functional plasticity, and postinjury behavioral experience is a major modulator of neurophysiologic and neuroanatomic changes that take place in the undamaged tissue.28 It has been found that task-oriented physical training could help the redevelopment of lost motor functions by organizing neuromuscular pathways with compensatory motor centers.28 Intensive repeated robot-assisted training has helped the functional motor recovery process for persons after stroke even during the chronic state.5 Fasoli et al9 indicated that poststroke motor recovery was similar to motor learning to some extent, and what was known about motor learning may predict the course of motor recovery. In motor learning studies, the learning of a skilled movement has been characterized by a plateau of little or no change in performance.29 Therefore, in this work, after the 10th session, the tracking skill had been stably developed or learned by the subjects after stroke, because the RMSE reached its steady state after the 10th session.

The changes in muscle coactivation patterns during the task-oriented and robot-assisted training were analyzed by the electromyographic activation levels of individual muscles and the cocontraction indexes of different muscle pairs. Before the 10th session, the overall RMSE values during tracking were relatively high. This was associated with the higher overall electromyographic levels of the biceps brachii, triceps brachii, and anterior deltoid muscles in these sessions (see fig 4), during which the cocontractions, that is, the overall cocontraction indexes, between the different muscle pairs were also high (see fig 5). Two major reasons could explain the higher electromyographic levels before the 10th session: the overactivation of muscles during the initial period of motor learning for a skillful task,30 and the spasticity after stroke, which could cause extra muscle activities.31 The high cocontraction levels in the muscle pairs observed during this period were also mainly associated with the excessive electromyographic activities of the muscles. The significant decreases in the overall electromyographic levels of the biceps brachii and triceps brachii, that is, the main agonist and antagonist muscles related to the elbow joint, occurred in the middle sessions of the training (from the 8th to 12th sessions), but the overall electromyographic levels of the 2 muscles were almost stable in the latter sessions (after the 12th session). This possibly implies that the reduction in the muscle spasticity, measured by the MAS after the training, mainly occurred in the middle sessions (ie, from the 8th to 12th sessions). The reduction in the electromyographic levels between an antagonist muscle pair also released the cocontraction between the muscles during elbow flexion and extension, illustrated by the cocontraction index values in the latter content. It is also understood that the reduction in electromyographic activation level during tracking was possibly because of the increased muscle force with resistive training, that is, the subjects could perform the tracking tasks with less muscle effort. However, the resistance applied to each session was proportional to the maximum flexion and extension torques of the same session, which was associated with the possible improvement in muscle force generation. Therefore, the reductions in electromyographic levels observed should be mainly attributed to the decrease in excessive muscle activations. The decreasing trend of the overall anterior deltoid electromyographic level suggests the reduction in the excessive muscle activations of the shoulder during the elbow tracking. Compared with the anterior deltoid muscle, the overall electromyographic level of the posterior deltoid muscle decreased little across the training sessions. This is possibly because flexors are usually more spastic than extensors in persons after stroke and muscle weakness is more commonly observed in extensors rather than flexors.25, 32

The cocontraction between the antagonist muscle pair around a joint in subjects without impairment could contribute to stabilizing of the joint in a static motion,33 and to movement accuracy in a dynamic motion.34 However, excessive cocontractions are energetically expensive,34 and abnormal muscle coactivation patterns, mainly concontractions, in the paretic limb after stroke even degrade the accuracy and efficiency of limb movements.15 The significant decreases in the cocontraction index values of all muscle pairs (see fig 5) was associated with the improvement in the tracking accuracy represented by the reduced RMSE (see fig 3). These decreases suggest overall improvement in the coordination of the individual muscles, or more effective and efficient muscle coordination for achieving the tracking task by the elbow flexion and extension. There are 2 major reasons that could explain the decreased cocontraction index of a muscle pair: the reduction in the electromyographic activation level of the muscles, and the reduced cocontracting phase of the 2 muscles. The decrease in the overall cocontraction index of the triceps brachii and posterior deltoid was mainly due to the reduction in the electromyographic level of the triceps brachii, because the variation of this cocontraction index was consistent with the change in the overall triceps brachii electromyographic levels. However, for the overall cocontraction index between the biceps brachii and triceps brachii, further decrease after the 10th session was observed when the overall electromyographic levels of the 2 muscles were almost unchanged. This suggests that the decrease was not related to the reduction in the excessive electromyographic activation of the antagonist muscle pairs, but was associated with the reduced cocontracting phase of the biceps brachii and triceps brachii. It also indicated an improved contracting and relaxing phasic pattern of the biceps brachii and triceps brachii during elbow extension and flexion in the latter sessions of the training. The decrease in the overall cocontraction index of the biceps brachii and triceps brachii appeared later, with a longer time course than the decreases in electromyographic levels of the respective biceps brachii and triceps brachii muscles. No steady state was reached by the decreasing cocontraction index of the biceps brachii and triceps brachii in the 20 sessions. It implied that further improvement in the muscle coordination between the biceps brachii and triceps brachii could possibly be obtained by providing additional sessions; and electromyographic cocontraction index, that is, cocontraction index, could be used as a simple measure for monitoring the improvement in muscle coordination during the training. For many persons after stroke, their elbow movements are usually associated with unnecessary shoulder activities. The decrease in the overall cocontraction indexes of the biceps brachii and anterior deltoid, triceps brachii and anterior deltoid, and biceps brachii and posterior deltoid after the 10th session indicated the better isolation of the elbow movement from the shoulder, which was related to the reduction in excessive activation of the anterior deltoid muscle. The decrease in the overall cocontraction indexes of the biceps brachii and anterior deltoid, anterior deltoid and posterior deltoid, and triceps brachii and anterior deltoid to the 20th session also suggests that further reduction in these cocontraction indexes was possible with more training sessions. It has been reported in existing literature that poststroke training for the distal joints could increase the motor capacity related to the intralimb proximal joint35 and the elbow training in this work might also benefit the functional recovery of the shoulder joint.

As reported by subjects in this study, the major improvement they all could feel after the training was the significant reduction in contracture and stiffness at the elbow and shoulder joints. This could be related to the great reduction in MAS score after the training (decreased almost half in the mean value compared with that of the pretraining) (see table 1). However, the improvements in FMA and MSS did not have as much change as in MAS after the training. Further improvement in motor functions assessed by FMA and MSS is possible with more training sessions in future studies, because the decrease in cocontraction index values did not reach a plateau after the 20 sessions of training in this study.

Conclusions 

In this study, significant motor improvements assessed by MAS, FMA, and MSS were observed after the 20-session training on elbow tracking task actively assisted by a rehabilitation robot. The muscle coactivation patterns during the interactive robot-assisted training on elbow flexion and extension were analyzed by the electromyographic activation level of individual muscles and the electromyographic cocontraction index of the muscle pairs. The electromyographic activation levels of the major agonist and antagonist muscle pair of the elbow joint, biceps brachii and triceps brachii, significantly decreased in the first half of the training course, which was associated with an improvement in tracking skill and a decrease in spasticity. The electromyographic level of the anterior deltoid also decreased during the training, suggesting a better isolation of elbow movements from the shoulder in the paretic limb. The coordination among the individual muscles related to the elbow and shoulder joints in the paretic upper limb improved mainly in the latter half of the training course, and was reflected in the reduction in the overall cocontraction indexes of the different muscle pairs. The results obtained in this study provide further understanding of the recovery process, especially muscle coordination, during interactive robot-assisted training, which would be useful for the design of robot-assisted training programs. We suggest that clinical studies with larger subject sample sizes to examine robot-assisted rehabilitation effects related to different training intensities could be carried out in the future studies.