[ARTICLE] Boosting robot-assisted rehabilitation of stroke hemiparesis by individualized selection of upper limb movements – a pilot study – Full Text

Abstract

Background

Intensive robot-assisted training of the upper limb after stroke can reduce motor impairment, even at the chronic stage. However, the effectiveness of practice for recovery depends on the selection of the practised movements. We hypothesized that rehabilitation can be optimized by selecting the movements to be practiced based on the trainee’s performance profile.

Methods

We present a novel principle (‘steepest gradients’) for performance-based selection of movements. The principle is based on mapping motor performance across a workspace and then selecting movements located at regions of the steepest transition between better and worse performance.

To assess the benefit of this principle we compared the effect of 15 sessions of robot-assisted reaching training on upper-limb motor impairment, between two groups of people who have moderate-to-severe chronic upper-limb hemiparesis due to stroke. The test group (N = 7) received steepest gradients-based training, iteratively selected according to the steepest gradients principle with weekly remapping, whereas the control group (N = 9) received a standard “centre-out” reaching training. Training intensity was identical.

Results

Both groups showed improvement in Fugl-Meyer upper-extremity scores (the primary outcome measure). Moreover, the test group showed significantly greater improvement (twofold) compared to control. The score remained elevated, on average, for at least 4 weeks although the additional benefit of the steepest-gradients -based training diminished relative to control.

Conclusions

This study provides a proof of concept for the superior benefit of performance-based selection of practiced movements in reducing upper-limb motor impairment due to stroke. This added benefit was most evident in the short term, suggesting that performance-based steepest-gradients training may be effective in increasing the rate of initial phase of practice-based recovery; we discuss how long-term retention may also be improved.

Background

Upper-limb (UL) motor impairment is a common outcome of stroke that can severely hamper basic daily living activities [1, 2, 3]. Training-based therapy can promote recovery with the outcome depending on the intensity and duration of the intervention [4, 5, 6]. Robot-assisted training allows intense practice without increasing the individual’s dependence on a therapist and can improve clinical scores of UL motor capacity [7, 8, 9]. However, the effects are usually small and provide limited improvement in motor function, especially in more severe hemiparesis [6, 7, 10, 11, 12]. Identifying training methods that can boost outcome is thus vital. Considering the extent of effort and sophistication invested in robot-assisted technology (e.g. [13, 14]) perhaps it is time to focus on how to optimise its utility (in terms of training principles). Recent attempts have focussed on creating training scenarios which are more engaging or which simulate daily living activities. However, the evidence for the added benefit of this approach is mixed [15]. Another approach is to individualize the difficulty of the practised task (e.g. [16, 17]). This is based on the idea that motor improvement depends on the ability to ‘make sense’ of information related to performance [18], and postulates that matching the challenge (difficulty) level of the training task to the current ability of the trainee would optimise motor learning [19]. Individualizing task difficulty is commonly achieved by adjusting the parameters controlling task demands (e.g. movement speed or distance; or amount of assistance) across a pre-selected standard set of movements, to match the ability of the individual. Yet, so far there is little evidence for the added benefit of this approach for UL motor recovery. Hence, individually adjusting the task difficulty level might –by itself – not suffice for boosting UL rehabilitation outcome.

We hypothesised instead that appropriate selection of the practiced movements – in terms of the muscle coordination patterns – is a key for improving motor recovery. UL hemiparesis can affect various aspects of control. Thus, different motor impairments may benefit from different training movements. For example, training with movements involving mainly patterns of intact muscle coordination is unlikely to contribute much to improve other impaired movement patterns, regardless of the task difficulty level. Similarly, training that focuses only on movements that involve severely impaired muscle control may contribute little, even if the task can be performed by compensatory movements. Hence, to be optimally effective, individualized training may need to be expressed, not only by individually adjusting the level of difficulty of the task, but also in selecting tasks which are relevant for recovery. Little has been done to explore this possibility (for some attempts see [20, 21]). Here we present a novel method for performance-based selection of the set of movement tasks for robot-assisted training. The method depends on the availability of a motor performance “map” that profiles performance across a workspace. Movements are selected within intermediate levels of performance, based on the variation of performance across the map. Specifically, we predicted that optimal reduction of UL hemiparesis would be achieved by training with movements located at points on the map of steep transition (steep gradient) from high to low performance (Fig. 1), thus promoting the cascade of generalisation of motor improvement. Improved performance of movements at these steep gradient locations on the performance map would steer improvement in neighbouring, but more impaired regions, and encourage recovery. Here, we present evidence supporting this hypothesis.

Fig. 1Illustrative sketch of the principle of selection of trained movements, based on the steepest gradients in a hypothetical motor performance profile (e.g. reaching aiming; vertical axis) measured across some particular task parameter (e.g. movement direction; horizontal axis); for simplicity, we show here a single dimension. The selected movements (grey horizontal bars) correspond to the regions with the steepest performance gradients, indicated by dashed ellipses. This movement selection principle can be applied where movement tasks can be defined by one or more continuous parameters, i.e. in a 1D, 2D, or higher dimensional map as long as the derivative of performance can be calculated. In this study we applied this principle on two measures of reaching performance (ability to move and ability to aim) each measured across two dimensions of the task (target location and movement direction)

To apply our method we first developed a novel principle of mapping of robot-assisted reaching performance across two dimensions of target location and movement direction [22], informing us about postural and movement-related aspects of motor control, respectively—key factors in the planning and execution of reaching movements [23, 24, 25]. The performance maps then served to select movement sets for training, based on our “steepest gradients” principle. To test our hypothesis–namely, training based on that principle would lead to superior recovery–we compared the outcome of 15 sessions of robot-assisted training between two groups of people who have severe-to-moderate chronic UL hemiparesis due to stroke, differing only in the selection of trained movement. In one group the selection was based on the steepest performance gradients principle (updated weekly) whereas the other group was trained with a fixed set of centre-out reaching movements regardless of participant’s performance profile, as commonly used in robot-assisted UL therapy [26].[…]

 

Continue —-> Boosting robot-assisted rehabilitation of stroke hemiparesis by individualized selection of upper limb movements – a pilot study | Journal of NeuroEngineering and Rehabilitation | Full Text

Fig. 2Experimental design. a. The sessions in each of the 3 participation phases are shown, with different colours indicating different session type. CA: clinical assessment; Map: mapping session. The first CA also served for screening. b. Schematic description of the experimental setting (top view; adapted from [32]). The participant held the robot handle, with grip ensured by a glove (Active Hands Co Ltd) and arm supported against gravity (SaeboMass, Saebo Inc.; not shown), which—at the beginning of each trial – was gently moved by the robot to a start position (white on-screen disc). Next, a target appeared on the horizontal display (blue on-screen disc; here shown black) and the participant tried to reach the target within the allotted time as accurately as possible, with the robot providing assisting and guiding forces as needed at each moment. Hand position was indicated on-screen by a red disc (not shown here). The horizontal display occluded the hand and the manipulandum from vision. Participants wore a harness to restrict trunk movement, keeping their forehead on a padded headrest attached to the workstation frame. The assistive force (Assist) promoted slower-than-allowed movements and also impeded very fast rebound-like movements characterising high elbow flexor muscle tone. The guiding force (Guide) impeded lateral deviation from a straight path towards the target. An animated ‘explosion’ was presented at the end of each trial with its final radius indicating reach accuracy (not shown). Also, during training sessions a 4-bar histogram summary, shown after each block (84 trials), informed the participant about his or her ability to initiate movements, move, aim and reach the target (adopted from [16]). c. The reaching workspace used for mapping performance. The locations of the 8 targets are indicated by small open circles and are specified by angular coordinates relative to the centre. An example of the hand located at the 90^otarget is shown. Participants made 5 cm reaches to each target from 8 start locations (indicated, for the example target, by small black dots and arrows), which were also specified in angular coordinates relative to the particular target. Note that the start coordinates therefore correspond to intended movement direction. The dashed circle indicates the extent of the mapped workspace, centred 24 cm in front of the headrest