Scientific Papers

A robot-aided visuomotor wrist training induces motor and proprioceptive learning that transfers to the untrained ipsilateral elbow | Journal of NeuroEngineering and Rehabilitation

Description of Image


Fifteen right-handed healthy adults (age: 23.5 yrs. ± 4.9 yrs.; 5 males) were recruited for this study. All participants gave written informed consent prior to testing. The study protocol was reviewed and approved by the Institutional Review Board at the University of Minnesota’s Human Research Protection Program. All participants reported no health problems and no known neurological conditions at the time of testing. Handedness of the participants was determined by the Edinburgh Handedness Inventory [16].


Wrist Robot: A three degree-of-freedom robotic exoskeleton (Fig. 1A) was used for the evaluation of wrist proprioceptive and motor function at pre-, posttest and retention as well as for the training sessions. It allows for motion within the full range of movement of the human wrist/forearm around its degrees-of-freedom (i.e., wrist flexion/extension, wrist adduction/abduction, and forearm supination/pronation). The wrist robot is a fully backdrivable system, powered by four brushless motors with the capability of delivering precise position and velocity stimuli to the wrist. The robot accurately encodes the wrist position at 200 Hz with a spatial resolution of 0.0075°. In addition, the robot is integrated with a virtual reality environment that in this study displayed the training task and provided instantaneous visual feedback of the user’s wrist position during the training session.

Fig. 1
figure 1

Experimental setup. (A) Frontal view of the 3 DOF robotic wrist exoskeleton. This study only required users to make wrist flexion/extension movements. (B) The elbow-joint manipulandum used for the transfer task. It allowed elbow flexion/extension movements in the horizontal plane. (C) Visual display as seen by the learner. Wrist flexion/extension movements tilted the virtual table. Learner attempted to roll the virtual ball into the target zone. (D) Learning effect on movement trajectory formation as measured by the Cumulative Spatial Error (CSE). Each data point represents the mean CSE of all participants for a particular trial. Note the decline of CSE over successive trials. Red line indicates the fit of the exponential decay function

Elbow joint manipulandum: A custom-built 2-joint manipulandum (see Fig. 1B) was used for assessing a possible transfer of proprioceptive and motor learning to the elbow. Technically, the manipulandum allows rotation around the wrist and elbow. Here, participants only performed elbow extension/flexion in the horizontal plane while the wrist joint movement was blocked during testing. A laser was attached to the front of the device to indicate the current arm position. Two US Digital H6 optical encoders (2500 quadrature count/revolution; spatial resolution: 0.036°), housed at the rotating point of the manipulandum lever arm segments, recorded angular position at a sampling frequency of 200 Hz.

Experimental protocol

The protocol comprised 4 sessions spread over three days: At Day 1, participants’ proprioceptive acuity and motor performance in a goal-directed pointing task were assessed at both wrist and elbow (pretest). On Day 2, participants completed the visuomotor training and the proprioceptive acuity and motor performance assessments were repeated (posttest). A final retention assessment was completed 24 h later on Day 3 (Fig. 2A). Prior to the pretest on Day 1, all participants underwent practice trials to get familiar with the devices and the tasks. The order of assessments (proprioception or motor) was randomized between subjects to account for possible order effects.

The evaluation of proprioceptive acuity consisted of a psychophysical assessment of wrist and elbow position sense using a two-alternative forced-choice discrimination paradigm. In each trial, the device passively moved the joint (right wrist or right elbow) at a constant velocity of 6°/s from the neutral position to either a standard (15° for the wrist, and 20° for the elbow) or a comparison position (> standard). Participants were then asked to verbally indicate which of the two positions was more flexed (see Fig. 2B). The subsequent comparison stimulus for the next trial was then determined by size difference between the two experienced stimuli and the previous verbal response using an adaptive Quest algorithm [17]. During testing participants wore opaque glasses to block vision and wore headphones that played low-volume pink noise to mask any auditory cues. The validity and reliability of this robot-based proprioceptive assessment have been established previously [18].

Fig. 2
figure 2

Schemata of experimental design and the procedure for position sense testing of the wrist. For assessing the elbow joint position sense, the procedure was analogous but used the elbow manipulandum (see Fig. 1B).

To evaluate untrained sensorimotor performance, participants performed goal-directed pointing movements matching with right wrist and right elbow in an ipsilateral matching task. Like during proprioceptive testing, vision and audition was blocked. In each trial, the respective joint (wrist or elbow) was passively rotated from the neutral joint position to a target position (15° wrist flexion, and 20° elbow flexion), held for 2 s and then moved back to the neutral position. Subsequently, participants actively moved their forearm or wrist to the previously perceived target position. Each assessment consisted of 20 trials.

The protocol for the visuomotor training was identical to and described in detail in our previous paper on contralateral transfer [15]. In brief, participants watched a visual display and had to move a virtual ball rolling on a tiltable board into a target zone by making continuous wrist flexion/extension movements (see Fig. 1C). A trial was considered to be completed upon holding the ball within the target zone for 5 s. Consecutively, a new target zone was presented to begin the next trial. If the trial was completed within 60 s, it was considered successful. Between successful trials, the wrist position corresponding to the horizontal position of the table (where the ball would be stationary) was altered to either 10°, 15° or 20° of wrist flexion (relative to the neutral joint position). This allowed for the training of several distinct wrist flexion positions within the available range of motion of the joint. It promoted sensory-based learning across workspace of the joint. Participants were not informed of these changes in the balancing position. After a participant completed at least one successful trial in each of the three different wrist positions, the task difficulty was automatically increased by altering the following virtual mechanical properties: (1) increasing the virtual mass of the ball and increasing the gain of the velocity of the virtual ball, and (2) decreasing the friction coefficients on the virtual table. Participants used a movement range of 10° wrist extension to 40° wrist flexion to complete the training trials. To prevent fatigue, the training session was limited to a maximum of 90 training trials or 45 min. At optimal performance (every trial was successful) a participant would have completed 30 levels of difficulty. Participants were allowed a 2-minute break after every 30 trials.

Measurements and statistical analysis

Evaluation of task-specific motor learning

To evaluate the effects of motor learning during the trained visuomotor task, the wrist angular time-series data of all participants were recorded during training using the signals of the position encoders of the robot. Instantaneous lateral deviation (LD) of the current wrist position relative to the neutral wrist position required to balance the ball was computed. In addition, Cumulative Spatial Error (CSE) for each trial was calculated using the equation below:

$$CS{E_{trial}} = \int_{i = 1}^n {\sum | } L{D_i}|dt$$

Where n is the last sample of each trial and trial onset (i = 1) is defined as the appearance of a new target zone. In addition, movement time (MT) of each trial was determined as the time difference between the appearance of a new target zone and the time when the virtual ball was held in the target zone for 5 s by the participant. Changes in CSE and MT represent measures of task-specific motor learning. Given that the trained virtual ball balancing task required learners to make continuous corrective movements until the ball was in the target zone, CSE reflects movements in both directions (flexion and extension).

Evaluation of proprioceptive acuity

Proprioceptive acuity was evaluated in the wrist position sense discrimination task described above. Participants’ verbal responses and the corresponding stimulus difference size (angular difference between comparison and standard positions) were recorded after every trial. Based on the verbal responses, a Just-Noticeable Difference (JND) threshold was determined by fitting the correct response rate and the stimulus difference size using a logistic Weibull function [19]. The JND or the marginal threshold x slope posterior distribution was derived by summating across the lapse rate dimension using the following psychometric function:

$${p}_{\alpha ,\beta }\left(\alpha =a, \beta =b\right)=\sum _{l}p(\alpha =a, \beta =b,\lambda =l)$$

where p(α = a, β = b, λ = l) is the full posterior distribution defined across the threshold values a, slope values b, and lapse rate l values that are contained within the parameter matrix defined a priori. The resulting JND threshold represented a measure of proprioceptive acuity.

Evaluation of untrained sensorimotor performance

Untrained sensorimotor performance was assessed by a discrete wrist or elbow pointing task described above. In each trial the participant’s joint was rotated to a target position. Subsequently, the participant actively rotated the hand or forearm to the previously experienced joint position. For each trial, the absolute angular error between the target position and the final joint position at the end of the active pointing movement was computed for each trial. A Movement Accuracy Error (MAE) was calculated as the absolute angular error between the target position and matching position across all trials for each participant using the equation below:

$$Movement\,Accuracy\,Error\,({\rm{MAE}})\, = \,\frac{{\sum\nolimits_{i = 1}^{20} {(|matching\,position – target\,position|)} }}{{20}}$$

Statistical analysis

Distributions for all variables were examined for normality using the Shapiro-Wilk test. Outliers were defined as data points falling 1.5 times above or below the interquartile range (IQR) (20). One outlier was identified at wrist MAE measures during the retention tests and was removed from further analysis. The remaining values for both wrist and elbow datasets were normally distributed, and parametric statistical analysis procedures were employed. To determine immediate training related differences in proprioceptive acuity and motor performance, paired t-tests were performed on the outcome measures JND and MAE for the right wrist and elbow at posttest relative to pretest. For the same measures, paired t-tests were performed at retention (24 h after practice) relative to pretest to examine if possible training effects were retained. The performed t-tests were one-tailed test as previous work (15) had already established that JND and MAE decrease as a function of training and would not induce a deterioration of proprioceptive acuity. The initial significance level was set at p = 0.05. To account for multiple testing, false discovery rate corrections using the Benjamini–Hochberg procedure were applied (21). All statistical comparisons were performed by using the Statistical Package for Social Sciences (SPSS) version 24.0. Effect size (Cohen’s d) and power calculations were computed using G*power 3.1.

Description of Image

Source link