Movement therapy without moving – first results on isometric movement training for post-stroke rehabilitation of arm function

Abstract

This study explores the use of isometric movement training for arm rehabilitation after stroke. The aim of this approach is to enhance movement skill even when the person training is not moving. This is accomplished by deceptively displaying virtual motions, exploiting known cross-modal sensory interactions between vision and proprioception. This approach can be advantageous in situations where actual movement is prohibitive due to weakness, spasticity, instability, or unsafe conditions. We present early insights on usability of and tolerance to this training approach and quantitative results that can power future clinical trials.

I. INTRODUCTION

Recent studies suggest that sensorimotor training with interactive technologies (e.g. robotics or virtual reality) can improve upper extremity function after stroke [1–3]. While many robotic devices are being used in large rehabilitation centers and many more prototypes are being developed in research laboratories, the large size, mechanical complexity, and cost of such devices are major drawbacks to their practical application in small clinics or at home.

In [4–6] we suggested that simple devices can foster motor learning of complex dynamics without the need for robotics or any other apparatus with complicated mechanical structures. Our approach is to exploit the nervous system’s natural capacity to adapt to different dynamic environments by using cross-modal sensory interactions. Our research and others have shown how the nervous system can be “tricked” by visually altering feedback [7–13], when disturbances are amplified [14, 15], or when feedback is falsified [16]. Such cross-modal sensory interactions have motivated training approaches for stroke survivors, such as visual error augmentation [17–19].

Here, we propose a movement therapy consisting of virtual movements in the absence of real movement – isometric movement training [4, 5]. During isometric training, a computational model of movement is used to make a graphic rendering of an avatar moving realistically in response to the measured interaction forces that subjects generate while constrained to a static position. Such graphic rendering aims at exploiting interactions between vision and proprioception to promote motor learning and to encourage users to believe they are moving, while they are not. Such approach can easily be implemented on a force measuring device without moving parts.

While isometric movement training appears to be a radical concept, this approach is backed by our previous studies showing that two-dimensional movements under a velocity-dependent force perturbation can be learned in a one-dimensional space [6, 20] or isometrically (i.e., without moving) [4, 5] by eliciting appropriate visual distortions in response to interaction forces. At the very least, isometric training will engage the patient and encourage coordinated neuronal activity. Recent research suggests that brain areas related to movement execution are also activated when movements are mentally rehearsed or observed [21], explaining why motor performance on a particular task is influenced by pure observation of another person executing that task [22–24]. There is some clinical evidence of movement observation as a potential tool in neurorehabilitation after stroke [25, 26]. Isometric movement training goes beyond motor observation or motor imagery, by encouraging subjects to also produce coordinated forces and muscle activity.

In this paper, we present a preliminary study of isometric movement training to promote arm function recovery in stroke survivors. While we do not expect clinically meaningful improvements to be observed in this early stage, the goal is to simply obtain early insights on usability of and tolerance to this training approach. This is a largely unexplored area in rehabilitation that could offer new benefits beyond conventional therapy.

II. METHODS

A. Subjects

Six stroke survivors participated in this proof-of-concept study. All subjects provided consent in accordance with Northwestern University and the Rehabilitation Institute of Chicago IRB (STU00104287). All participants (Table I) were naïve to the experimental procedures. To be included in this study, participants had to be at least 21 years of age and more than 6 months since stroke onset. Exclusion criteria were severe deficits in cognition or comprehension that would preclude participation in the intervention, neurological deficits not related to stroke, non-corrected visual deficits and neglect.

Table I –

Participants

Subject	Age	Gender	Side affected	Handed- ness	Months post-stroke	FMUE* score
S1	54	F	R	R	68	21
S2	55	F	R	R	42.5	30
S3	59	M	R	R	73.5	46
S4	64	M	R	R	109.5	27
S5	41	M	R	R	26.5	36
S6	66	M	R	R	90.5	26
Mean (SD)	56.5 (9.9)	-	-	-	68 (30.4)	31 (7.38)

^*FMUE – Fugl-Meyer Upper Extremity

B. Experimental setup

Participants were seated in a chair of adjustable height to accommodate individual requirements. Their impaired hand, with forearm in pronation, was attached to a 2-DOF manipulandum that constrained movements to the horizontal plane at the sternum level (xiphoid process). A support prevented wrist movements, and an arm support attached to the subject’s forearm kept their arm in horizontal position and cancelled the influence of gravity, to help increase arm range of motion [27]. A 2-DOF force sensor (Assurance Technologies, Inc., Bartlett, IL, USA) was located at the point of attachment between robot’s endpoint and the handle, directly below the palm of the hand.

A horizontal screen occluded the participant’s view of their own arm. A realistic graphic rendering of an arm (avatar arm), with subject-specific anatomical characteristics (i.e., arm length and position of centers of rotation relative to the subjects’ seat position) was displayed on the screen (Figure 1a).

Figure 1.

Schematic of the experimental setup. a) During isometric movement training an avatar arm moves based on forces applied by the subject, while the subject’s true arm remains stationary. b) During free motion condition, the avatar arm always overlaps with the subject’s true arm motions.

Participants could move the avatar arm in two different ways: free motion and isometric motion. During free motion, subjects could move their arm in the horizontal plane and the avatar arm matched the actual arm movements (Figure 1b). During isometric motion, subjects were constrained to the central position (hand position located along the midline at ~30cm away from the body) by clamping the manipulandum’s end-effector with a physical brake. Forces that subjects applied against the physical brake were used to move the avatar arm. The dynamics of the avatar arm were computed as in [4, 5] without considering the biomechanical deficits of the patients.

C. Experimental protocol

The experiment consisted of four phases: practice (free motion), pre-training (free motion), training (isometric motion), and post-training (free motion). The number of movements performed within each of these phases is summarized in Table II.

Table II –

Total number of movements (trials) per subject and experimental condition.

Subject	Practice (fm)	Pre- training (fm)	Isometric training (im)	Post- training (fm)	Notes
S1	few minutes	15	90	15	preliminary protocol (less trials, uncontrolled practice; no virtual damping inside the target during isometric training)
S2	15	15	105	15	preliminary protocol II (same trials as final protocol; no virtual damping inside the target during isometric training)
S3…S6	15	15	105	15	final protocol (virtual damping was added inside the target during isometric training to help subjects stabilize after they had successfully reach the target)

fm – free motion; im – isometric motion

During the experiment, a 0.5 centimeter diameter cursor was overlaid on the top of the avatar’s hand. Participants were required to make center-out movements of 15 cm towards three 1.5 cm diameter targets oriented at 120° from each other. Targets appeared in random order after completion of a movement (trial). After each reaching movement, the targets would change color. They would become green if movement duration was between 1.2 s and 1.8 s, blue if movement time was above 1.8 s and red if it was below 1.2 s. This timing was heuristically proven comfortable for post-stroke subjects that participated in earlier experiments in our lab.

The movement duration feedback was implemented to ensure the average speeds of movement remained relatively constant. After a stabilization period of 0.5s, inside the target, subjects were asked to return to the central position and wait for the next target. If participants could not perform the required center-out movement within 5 s, the target would disappear and the subject would be asked to come back to the central position.

Subjects were assisted during the movement returning to the central position by means of a virtual spring-damper element connecting the avatar’s hand with the central position.

At the end of the experiment, participants were asked to give feedback to provide further information about acceptability of the protocol. Four participants were asked to rate statements about usability, tolerance, and impressions regarding the isometric training on an integer scale between 1 and 7, were 1 meant that the assertion was not at all true and 7 very true.

D. Safety measures

Subjects were allowed to interrupt the experiment at any time and for any length of time between trials, and to terminate or suspend the experiment. The experimenter was always monitoring any possible adverse events, or signs of increased pain or fatigue. If any of those events happened, the experimenter would stop the experiment immediately.

E. Data analysis

We evaluated the straightness and smoothness of the movement trajectories. For straightness, we quantified directional error as in [4], which can be interpreted as an indicator of the “straightness” that resulted from feedforward control commands, and total error, which represents the total lateral deviation of the movement trajectory relative to the line connecting the center position and the target. For smoothness, we computed the spectral arc-length of the movement speed profile, SPARC [28] and the logarithm of dimensionless jerk, LDJ [29, 30].

For intra-subject comparisons, Mann-Whitney tests were performed. For group analysis, sign tests were performed. Outliers, defined as those outcome measures beyond 1.5 times the interquartile range of each experimental phase, were removed from the data sets prior to any statistical tests.

III. RESULTS

All the subjects tolerated and completed the protocol, and no adverse events were recorded during the experiments. There was no report of pain, discomfort, fatigue, or other negative symptoms. All subjects but one felt as if their arm was moving while observing the virtual arm with their own arm locked. Subjects felt it was easier to move the virtual arm than it is to move their affected arm. They found the isometric training enjoyable, fun, and interesting. They described the activity as useful, important, and that it could be beneficial to them. They did not feel bored, anxious, tense, pressured, or have difficulty paying attention during isometric training. Finally, they reported that they were confident that they performed the activity well.

We found no significant improvements in our performance metrics (pre- vs. post- training), but also no unwanted effects. While both real (Figure 2, blue) and avatar (Figure 2, green) motions were attempted, not all actions were complete (Figure 2, red). As a group, participants moved equally straight before and after isometric training in the free motion conditions. While movements seemed to be less smooth after isometric training than before training as measured by SPARC, this observation was not detected by the LDJ measure (Figure 3).

Figure 2.

Left. Cursor trajectories of post-stroke participants. Subjects performed 15cm long point-to-point reaching movements in isometric (green) and free motion (blue) conditions. Red trajectories correspond to unsuccessful trials, in which the target was not reached in 5 seconds. Right. Box plots of the different outcome measures (without outliers). Each box represents the data from 15 trials (5 trials per direction) and represents the two quartiles surrounding the median, and lower and upper boundary lines indicate 1.5 IQR from the 1^st (3^rd) quartile or the maximum (minimum) value, whichever was closest to the median. Lines between whisker boxes denote statistical significance (Mann-Whitney test, p<0.05).

Figure 3.

Group results - Each marker represents the median value for each subject (without considering outliers). The spread of the whisker boxes were computed as in Figure 1. Asterisks denote statistical significance (sign test, p<0.05).

Subject 1 (S1) was unable to reach one of the targets in free motion, but successfully reached all targets during isometric training. The smoothness of the virtual movements during isometric training seemed to improve (SPARC p<0.05; LDJ not statistically significant, p=0.197, but showed a tendency towards improvement).

Subject 2 (S2) was unable to reach one of the targets before training, but successfully completed all reaches to that target after training. S2 initially struggled with controlling the avatar, but improved significantly throughout training as indicated by the reduced number of unsuccessful trials, and improvements in the straightness and smoothness of virtual movements (total error p<0.001; SPARC p<0.05; LDJ p<0.01).

Subjects 3–6 (S3-S6) did not improve straightness or smoothness after training. However, qualitatively, these subjects seemed to improve in controlling the avatar during isometric training. Outcome measures of virtual movements at the beginning and end of isometric training were not statistically different (except S6 total error p<0.01). However, subjects reduced the number of unsuccessful trials, implying a better control of the avatar.

IV. DISCUSSION

This work provided very preliminary information on the prospects of isometric movement training [4, 5] as movement therapy for stroke survivors. We presented initial evidence of the safety and feasibility of this novel rehabilitation technique to improve upper-limb function after a stroke. The study was not designed to test clinical efficacy; the goal was simply to obtain early insights on usability and tolerance of this training approach. To this end, we gathered encouraging feedback and found encouraging results that can guide and power future studies and clinical trials.

No adverse events were found and subjective reports to assess participants’ impressions about the isometric training approach were promising. Participants liked the approach, found it was easy to engage with, and that it was an activity that could help them. They also declared that most people with similar condition to them would learn to use this system very quickly.

The isometric training suggested here has the potential to engage the visuo-motor coupling system and motivation in an intervention that requires subjects to produce coordinated forces and muscle activity. This approach can be significant because, if proven effective, it has a broad range of potential benefits: i) it will allow motor skill training in situations where movement is prohibitive, such as in the presence of weakness, spasticity, instability, or unsafe conditions; and ii) it will enable training devices that are safer, cheaper, and more accessible than current robotic approaches, thanks to the reduction in the training device’s mechanical complexity.

In conclusion, isometric training can be a simple, yet promising approach to treat patients with different levels of upper-limb impairments after a stroke. It also seems to be safe for further study.

Finally, further studies could investigate this approach in other neurological populations, such as cerebral palsy, and use of electromyography (EMG) to consider the effects of muscle co-activation on impedance modulation during (virtual) movements.