Background
Many upper-limb amputees rely on prosthetic hand devices to restore a degree of functionality to the performance of daily activities. Despite the increasing sophistication of these devices, they still provide less than 50% of the capability of an intact limb [
1,
2], impose a high cognitive burden that results in fatigue and frustration [
3], and are therefore frequently rejected [
4]. The nature of this cognitive burden has recently been explored indirectly by examining disruption to visuomotor behaviours during prosthetic hand use [
5,
6]. For example, Parr et al. [
7] showed that when using a myoelectric prosthetic hand simulator, participants directed a greater amount of visual attention towards the prosthesis and objects being manipulated by it. This dependency on visual feedback to monitor and correct movements is in contrast to the feed-forward (target-focused) strategy revealed by skilled users in everyday tasks [
8], and mirrors findings from novices in other domains (e.g. tool use [
9] and laparoscopic surgery [
10,
11]). Interestingly, it is this need to constantly, and consciously, pay close visual attention to movements that prosthesis users report as a key contributor to the cognitive burden experienced during prosthetic hand control [
4,
12,
13]. The overall aim of this paper was to assess novel measures of this cognitive burden and to test the efficacy of a novel training technique that might reduce this burden.
Measures that directly evaluate this cognitive burden are needed in order to further our understanding of how efficient visuomotor behaviour is influenced by prosthesis use. Electroencephalography (EEG) is ideally suited for this purpose as it offers a window into the dynamics of ongoing neural activity with high temporal resolution. This is important, as the development of skilled motor performance is characterised by the precise allocation of processing resources to areas of the brain that are needed for successful task execution; termed ‘neural efficiency’ [
14,
15]. It has been suggested that neural efficiency can be operationalised by cortical oscillations in the alpha frequency (8-12 Hz) [
16]. Specifically, the magnitude (power) of alpha oscillations influence cortical activation by exerting inhibitory control and can therefore reveal a gating mechanism whereby resources are diverted away from regions showing higher alpha power (more inhibition) and towards regions showing lower alpha power (lower inhibition) [
17]. Such a mechanism is reflected in evidence suggesting that during movement planning and execution, alpha power decreases over motor-related areas of the cortex while increasing over non-motor areas [
18].
Using this gating model, research has shown that enhanced performance in motor tasks can be characterised by more efficient topographical alpha power distributions. For example, Gallicchio and colleagues have shown that lower central alpha power and higher temporal alpha power preceded improved performance in a biathlon shooting task [
19] and were evident following a training period in golf-putting [
20,
21]. Indeed, higher alpha power over the left-temporal region has been generally associated with improvements in motor learning and performance [
22,
23], as conscious, verbal-analytical processes diminish as a function of automaticity and expertise [
14,
24‐
27]. It is therefore plausible to assume that the cognitive burden experienced during initial prosthesis hand control is underpinned by both neural inefficiency, a dependence on vision to monitor hand state and that both may reflect a more conscious mode of prosthesis control.
Discussion
This study provides the first direct examination of the cognitive burden associated with prosthetic hand control. As predicted, participants performed significantly (~ 4 times) slower when using the prosthesis simulator compared to their anatomical hand. Furthermore, this performance decrement was underpinned by spatial and temporal disruptions to hand-eye coordination. In line with Parr et al. [
7], participants exhibited significantly lower TLS (more hand-focused gaze) and significant delays in the time to disengage from hand movements in all phases of the task. This again supports the idea that novice prosthetic hand use is reflected by an increased dependence on vision to monitor hand movements [
5‐
7] and the inability to fixate targets ahead of time [
7]. As hypothesised, the phase of the task that required the highest dependence on vision was the Lift phase [
7]. During this phase, participants dedicated considerably more visual attention to the hand than the target (Mean TLS = − 49%) and took ~ 600 ms to disengage gaze from the jar following its pick up (the first 30% of the entire Lift phase).
When examining regional alpha power, our results revealed a focal pattern in which neural resources were directed away from occipital and temporal regions (generally highest alpha power) and diverted towards central and parietal regions (generally lowest alpha power), a pattern that was insensitive to both hand condition and movement phase. This pattern is in line with the gating-by-inhibition hypothesis [
17] and supports research evidencing the bilateral activation of sensorimotor processes required to perform reaching and grasping movements [
36]. It was surprising that this gating pattern was insensitive to hand condition given previous research has shown specific regional changes that occur as a function of expertise [
14] and learning [
31]. This is particularly the case for the left-temporal region that is thought to represent the conscious verbal processes present in the early stages of learning. However, such an effect may have been masked by the global decrease in alpha power that occurred during the prosthetic hand condition. Indeed, previous research has shown that novice performers exhibit a greater decrease in global alpha power compared to experts in visuomotor tasks [
16,
25,
26], reflecting the increased cortical activation and mental effort required to perform the task [
25]. Our results therefore support the hypothesis that initial prosthetic hand is underpinned by decreased neural efficiency as well as an increased dependence on vision. Examination of global alpha power could therefore provide a measure of skill development or cognitive effort to compliment measures of gaze in future studies.
However, contrary to our hypotheses, alpha power was consistent across both phases of our task despite these phases requiring distinctly target focused (Reach) and hand focused (Lift) visual strategies. This suggests that the cognitive processes behind visual attention are not straightforward, and raises questions concerning the validity of inferring the cognitive burden imposed during prosthetic hand control from overt visual attention alone [
6]. It is also possible that alpha power may not be a suitable measure to detect more subtle changes in cognitive functioning that develop throughout a task. Indeed, the link between alpha power and neural efficiency in motor tasks has primarily been based on expert-novice differences [
14,
15,
25]. Based on these considerations, regional alpha power may be more suited to reflect more radical or long-term changes in the functional architecture of the brain.
While these results are exciting, and could be used to quantify the usability and embodiment of prosthetic devices, questions remain concerning whether this cognitive burden can ever be alleviated, and, if so, which training interventions would be best suited to facilitate this process. Here, we have established that initial prosthetic hand control disrupts performance, increases the dependence on vision, and decreases neural efficiency. An interesting question going forward is whether training a prosthesis user to use their eyes more effectively would increase neural efficiency and facilitate the acquisition of prosthetic hand control. In the next experiment, we attempt to answer these questions by examining the impact of a gaze training (GT) intervention on measures of neural efficiency, conscious control and prosthetic hand learning.
Experiment 2
While there are no evidence based guidelines for teaching prosthesis use, instructions are generally very explicit in nature, focusing the patient’s attention on limb movement [
37]. Such instruction encourages the accrual of declarative knowledge and the conscious control of movement that can place high demands on attentional resources [
38]. This type of movement control is indicative of the early stages of learning where cognitive demands are high, performance is error strewn and vision is the dominant sensory modality used to supervise on-going action [
24]. In contrast, GT interventions use observational learning principles to guide novice performers to adopt eye-movement behaviours that are indicative of experts. Not only has GT been shown to expedite skill acquisition in novices learning surgical skills [
11,
33,
39], in patients with movement coordination disorders [
40‐
43] and in sports performers [
44‐
46], but this learning has been found to be more implicit [
34], and less cognitively demanding [
39] when compared to technical instructions focused on limb movements. GT may therefore prove fruitful for prosthetic hand rehabilitation by lowering demands on visual attention and potentially reducing conscious cognitive control.
A method of measuring conscious control is through EEG connectivity; the phase synchrony or “co-activation” between two signals from the brain, with high connectivity reflecting functional communication and low connectivity reflecting regional independence [
47]. Increased conscious movement control can be reflected by increased high-alpha (10–12 Hz) connectivity between the motor planning (Fz) and verbal-analytical (T7) regions of the brain [
48]. For example, T7-Fz connectivity has been shown to reduce as a function of expertise [
14,
20], and increase in individuals who are exposed to explicit rather than implicit training instructions [
48,
49], whereas connectivity between motor planning (Fz) and visuo-spatial (T8) regions are not as susceptible to change [
50]. Indeed, these disparate connectivity patterns have been shown in various skills, including surgery [
49], postural control [
51], rifle shooting [
14], and golf putting [
20,
50].
As well as providing a novel method of testing the efficacy of GT, EEG connectivity can allow further investigation into the relationship between visual attention and neural efficiency. Whilst topographical alpha power may reveal more long term changes in the functional architecture of the brain that arise via practice, evidence has shown T7-Fz connectivity to actively change in response to the ongoing context of practice; such as implicit vs explicit learning [
49], internal vs external focus of attention [
52], and increased task difficulty [
51]. In fact, Ghasemian et al. [
53] showed direct evidence that changes in EEG connectivity are sensitive to both short-term (same day) and long-term (1 week) training, whereas changes in EEG power are more affected by long-term changes. Therefore, alpha connectivity may be better suited to reflect a more immediate link between visually guided and consciously controlled movement than alpha power.
In this second experiment, we examined the efficacy of a GT intervention on prosthetic hand skill learning and retention compared to movement-related instructions typical of rehabilitation settings. Using a coin lifting task, we specifically focussed on the cortical dynamics occurring during object manipulation when demands on visual attention were highest. By doing so, we can clearly demonstrate how preventing learners from monitoring the prosthetic hand subsequently influences neural efficiency and learning. We also examined how effectively participants could transfer these skills to a more complex tea-making task. Accordingly, we make several hypotheses. First, we hypothesise that both interventions will facilitate performance improvements that should subsequently reduce the cognitive demands of the task. Second, we hypothesise that optimising gaze control (increased TLS & reduced gaze shifting) via GT will expedite learning and develop visuomotor strategies that are ultimately more neurally efficient (increased alpha) and less consciously controlled (reduced T7-Fz connectivity) compared to movement training (MT). As such, we expect a relationship between visual attention and conscious movement control to emerge. Finally, we hypothesise these benefits will be transferred to the more complex tea-making task.
Discussion
The aim of the second experiment was to determine the efficacy of GT in expediting prosthetic hand learning and alleviating the associated cognitive burden. We hypothesised that GT would optimise visual control, expedite skill acquisition, and promote neural efficiency by reducing conscious control, compared to MT instructions. We also hypothesised that these benefits would carry over to our complex transfer task [
33]. Finally, we hypothesised that an increased dependence on vision to monitor the prosthesis would be related to increases in conscious movement control.
Supporting our hypothesis, results suggest that participants in the GT group implemented the training instructions by increasing their TLS and increasing the speed of their gaze shifts compared to the MT group (Fig.
4). Our results also show that by adopting more efficient gaze strategies, participants in the GT group performed consistently faster than the MT group from the first training session onwards. Although both groups exhibited a significant improvement in performance across time that somewhat plateaued by the third training session, the natural speed of participants in the GT group was ~ 20% faster than the MT group without being any more errorful. Encouragingly, the improved visual control adopted by the GT group also transferred to the more complex tea-making task, with participants using a higher TLS (~ 20%) compared to the MT group (Fig.
7).
While GT optimised gaze behaviour and expedited learning, we found mixed results when determining whether this decreased dependence on vision enhanced neural efficiency. For regional alpha power, we found a focal pattern consistent with Experiment 1, in which cognitive resources were primarily gated towards the central and parietal regions of the brain, regardless of training received. As such, our findings seemingly validate the utility of measuring regional alpha power to examine the functional architecture of the brain during prosthetic hand control. Although this gating pattern was insensitive to change from baseline to retention, there was a significant decrease in temporal alpha at delayed retention compared to baseline, regardless of which training was received. This increased excitability of the temporal regions is contrary to our predictions that increased skill would decrease (left) temporal activity. However, it should be noted that our predictions were primarily based upon research comparing expert-novice differences (as was seen in Experiment 1), or longitudinal training (~ 15 weeks) in target sports. Given the dynamic nature and complexity of our task, it is likely that the putative link between motor-skill expertise and optimal cortical organisation, as indexed by alpha
power, might flexibly depend on external demands and required performance rather than a rigid strategy (always reduced activity [
56];). Future research could explore this by conducting longitudinal intervention studies or by examining expert vs. novice comparisons of prosthesis users.
Our EEG results did however provide stronger evidence to suggest that GT reduces conscious verbal-analytical processes. Specifically, we showed that participants in the GT group exhibited a significant reduction in T7-Fz connectivity from baseline to retention and delayed retention, whereas the MT group did not (Fig.
5). We also showed a significant difference in the baseline change in T7-Fz connectivity between groups, with GT showing a decrease and the MT group showing an increase. The change in temporal-frontal connectivity also showed significant hemispheric asymmetry for the GT group, showing decreased T7-Fz and increased T8-Fz connectivity. Encouragingly, similar results were observed in the transfer tea-making task, with the training conditions again significantly altering the change in T7-Fz connectivity. However here, participants in the GT group displayed similar levels to that seen at baseline coin task performance, whereas the MT group showed a large increase.
These findings strongly suggest that encouraging learners to engage visual attention on the target rather than object manipulation, discourages burdensome verbal-analytical control [
34]. In fact, regression analyses provided direct support for this claim, revealing that reduced T7-Fz connectivity was significantly predicted by increased TLS and faster gaze shifting times at retention and delayed retention. Conversely, our results also highlight how the provision of explicit instructions can accentuate the reliance on verbal processes, especially during complex tasks that are more reflective of the activities of daily living. Indeed, as these relationships were not present at baseline, the link between visual monitoring and conscious control appears to be highly dependent on the cognitive strategies encouraged through training rather than being inherent in prosthesis control. As conscious control processes require high cognitive demands they can result in performance breakdown under increasing task difficulty and fatigue [
38] and should therefore be minimised in prosthesis rehabilitation.
These results provide evidence that GT alleviates conscious control and promotes neural efficiency, reducing the non-essential interaction between the motor planning and verbal-analytical regions of the brain. They also provide evidence that the provision of explicit instruction via MT can have the opposite effect, increasing the functional communication between motor-planning and verbal-analytical regions – an effect that increased during the more complex transfer task. Indeed, these findings are in line with previous research in laparoscopic surgery [
49], and should not only act to promote the benefits of implicit learning via GT, but also act as a warning against the provision of more explicit training methods.
General discussion
In this study, we report the first attempt to simultaneously examine the visuomotor and cortical mechanisms that contribute to the cognitive burden experienced by upper-limb prosthesis users [
4,
13]. In both experiments, we provide further evidence that prosthetic hand control places high demands on visual attention and cognitive processes in order to guide and monitor movements, particularly during object manipulations [
7]. Importantly, we also show that individuals can be trained to reduce their reliance on vision via GT, which subsequently expedites learning and encourages greater neural efficiency compared to more traditional explicit training methods. The findings of these experiments therefore have important theoretical and practical implications.
From a theoretical perspective, it is important to understand why prosthesis users appear to maintain these inefficient strategies despite skill improvements. In the development of eye-hand coordination, vision is initially utilised primarily as a feedback mechanism to monitor ongoing action as learners develop sensorimotor mapping rules between commands and movements, and between vision and proprioception [
9]. However, for
typical learners, as these mappings are refined, dependence of vision is relinquished from monitoring action and begins to be used as a feed-forward mechanism as soon as other senses (primarily touch and proprioception) can take over [
8]. Prosthesis users’ over-reliance on visual feedback therefore represents a sensory substitution that is required to compensate for the severe deficits in proprioceptive and haptic feedback in this mapping process. Yet, considering the ease at which participants were trained to stop looking at the prosthesis in Experiment 2, it is clear that this strategy is not efficient nor a prerequisite of successful prosthesis control. So how then does GT help to overcome such deficits? And why might this be beneficial to long-term prosthesis control?
There are a number of potential theoretical explanations for this. First, being trained to use vision in this more proactive manner and to “
look at the right place at the right time” is thought to aid effective coordination of the visuomotor system [
8,
9,
43]. Specifically, by adopting early and accurate look-ahead fixations users are able to effectively pass visually acquired target-related information to the motor system so accurate movements can be planned and executed [
8,
9]. The faster performance times exhibited by the GT group support these predictions, and suggest increased proficiency of movements. Second, reducing the dependence on vision reduces conscious movement control, supporting the idea that GT alleviates the reliance on these explicit and burdensome processes [
34]. Third, it could be that case that GT forces the development of ‘new’ sensorimotor mapping rules using the remaining senses (e.g., proprioception, or auditory information from the prosthesis’ motors [
13]) to enable vision to be used in a more proactive feed-forward manner.
2 Finally, the benefits of GT could also be attributable to encouraging learners to adopt an external focus of attention (FOA). Research has shown that focusing on the effect of movement (external FOA) rather than the mechanics of the movement itself (internal FOA) promotes better performance in a variety of movement contexts [
57]. Interestingly, an external FOA has also been shown to improve movement economy by reducing muscle stiffness and activity [
58]. Reducing demands on muscle fibre recruitment may therefore mitigate the negative effects of fatigue upon electromyographic (EMG) signal quality [
58] and improve long-term myoelectric control.
From an applied perspective, the methods used in these experiments could be used to assess the usability of prosthetic hands from a design perspective. While the technological development of hand prosthesis is increasing rapidly, research examining the usability and interaction between the user and the prosthesis is lacking. For example, while performance measurements are adequate in accessing the functionality of prosthesis hand devices, they are not sensitive enough to assess their usability. As our transfer task shows, both training groups performed similarly but the magnitude of mental resources needed to perform was significantly less in the GT group. So, just because a user can
use a hand prosthesis does not mean that the hand prosthesis is intuitively
useable. From technologies that provide vibrotactile feedback [
59] to hands that can actually ‘see’ for themselves [
60], each will increase or lessen the cognitive resources needed to interact with the world. It is this user-prosthesis-world interaction that needs examining in future research, which to be effective, will depend on significant collaborations between applied psychologists, prosthesis engineers, occupational therapists and prosthesis users themselves.
Similarly, an examination of the cognitive demand experienced during prosthesis learning could also aid occupational therapists to assess a patient’s progress. However, the methods used in these studies are probably not cost effective given the expensive equipment required and the expertise needed to operate it. Researchers therefore should develop and validate a multidimensional workload measure specific to prosthesis use. Such a measure has previously been developed for surgical skills (SURG-TLX; [
61], and would allow for more cost-effective and immediate clinical assessment of the cognitive demand experienced by prosthesis users during the rehabilitation process.
Despite the important first steps presented here, several limitations should be noted. First, we are limited by our use of intact users of a simulator rather than patients with limb loss. However, evidence has shown that these populations display comparable kinematic profiles [
1], visuomotor behaviours [
6,
7], and perceptual experiences [
62], suggesting that using a simulator provides a useful surrogate to examine the sensory-motor deficits that prosthesis users face. Yet, it is unclear how increasing the length of the operating arm when using the prosthesis simulator (approximately 7 cm when the hand is unclenched) independently influences visuomotor and neurophysiological behaviours. Furthermore, the cortical reorganisation that occurs following amputation can cause large-scale changes in neural networks, making direct transfer of our results to an amputee population potentially difficult. For example, evidence shows that neuroplasticity of the cortex following amputation can promote an expansion of the residual limb segments into the former limb territory [
63], and promote a progressive disconnection of the missing hand cortex and the sensorimotor cortex [
64]. Clearly, future work is needed to evaluate the cognitive burden in a clinical population and to explore if this can be alleviated in the same manner using a GT intervention.
The degree of ambiguity in the temporal accuracy of EEG data must also be highlighted. Here, data were segmented through clearly defined epoch lengths relative to a given manual action (i.e., jar lift in experiment 1). Whilst this method enabled meaningful comparisons to be made, it fails to guarantee that the segmented data represent the exact same “portion” of movement on a trial-to-trial basis. Though unfavourable, this inaccuracy appears a necessary compromise for investigating EEG during dynamic motor tasks, that should be addressed in future research.
Finally, in these studies we limited our EEG analysis to the alpha frequency band in order to contextualise our findings with previous research on alpha gating [
31] and connectivity during movement execution [
20,
49]. In future, more exploratory research could benefit from investigating multi-scale interactions across different frequencies in order to acknowledge the fact that changes in specific frequency bands do not occur in isolation [
65]. Such analyses could help to attain a more holistic understanding of the cortical disruptions evident during initial hand use and this could help develop objective methods to assess training programmes in the future.