Visual attention, EEG alpha power and T7-Fz connectivity are implicated in prosthetic hand control and can be optimized through gaze training

Abstract

Background: Prosthetic hands impose a high cognitive burden on the user that often results in fatigue, frustration and prosthesis rejection. However, efforts to directly measure this burden are sparse and little is known about the mechanisms behind it. There is also a lack of evidence-based training interventions designed to improve prosthesis hand control and reduce the mental effort required to use them. In two experiments, we provide the first direct evaluation of this cognitive burden using measurements of EEG and eye-tracking (Experiment 1), and then explore how a novel visuomotor intervention (gaze training; GT) might alleviate it (Experiment 2).

Methods: In Experiment 1, able-bodied participants (n = 20) lifted and moved a jar, first using their anatomical hand and then using a myoelectric prosthetic hand simulator. In experiment 2, a GT group (n = 12) and a movement training (MT) group (n = 12) trained with the prosthetic hand simulator over three one hour sessions in a picking up coins task, before returning for retention, delayed retention and transfer tests. The GT group received instruction regarding how to use their eyes effectively, while the MT group received movement-related instruction typical in rehabilitation.

Results: Experiment 1 revealed that when using the prosthetic hand, participants performed worse, exhibited spatial and temporal disruptions to visual attention, and exhibited a global decrease in EEG alpha power (8-12 Hz), suggesting increased cognitive effort. Experiment 2 showed that GT was the more effective method for expediting prosthesis learning, optimising visual attention, and lowering conscious control - as indexed by reduced T7-Fz connectivity. Whilst the MT group improved performance, they did not reduce hand-focused visual attention and showed increased conscious movement control. The superior benefits of GT transferred to a more complex tea-making task.

Conclusions: These experiments quantify the visual and cortical mechanisms relating to the cognitive burden experienced during prosthetic hand control. They also evidence the efficacy of a GT intervention that alleviated this burden and promoted better learning and transfer, compared to typical rehabilitation instructions. These findings have theoretical and practical implications for prosthesis rehabilitation, the development of emerging prosthesis technologies and for the general understanding of human-tool interactions.

Keywords: Amputees; Conscious control; Inter site phase clustering; Intervention; Motor learning; Myoelectric prosthesis; Therapy.

Fig. 1

The myoelectric prosthetic hand simulator and the AOIs for the jar task. The prosthetic-hand simulator (a) and a screenshot taken from the Eyevision software (b). The screenshot shows the task environment and the 6 AOIs (1 = jar, 2 = carton, 3 = target, 4 = button, 5 = prosthesis, 6 = hand mat)

Fig. 2

Gaze and EEG data for the jar task. Mean (± SD) target locking scores (a) for the anatomic and prosthetic hand simulator conditions across the two phases of the task. Positive scores reflects more time spent looking at targets and negative scores reflect more time looking at the hand. Mean (± SD) time in milliseconds to shift gaze (b) for the anatomic and prosthetic hand simulator conditions across the two movement phases. Positive times reflect a gaze shift after completion of a task phase whereas a negative time reflects a gaze shift prior to the completion of the task phase. Scalp topoplots (c) representing the global distribution of alpha power across hand conditions. Line plot (d) representing alpha power (± s.e.m) recorded from each region of interest (ROI) for both the anatomic and prosthetic hand condition. As there was no effect of task-phase presented values for both (a) and (b) represent the average of the two phases (reach and lift) for each ROI

Fig. 3

Experimental setup and AOIs for the coin task and the transfer tea-making task. The experimental set-up for our modified coin task (left), annotated with our four AOIs (1 = jar, 2 = coin (× 4), 3 = prosthesis, 4 = drag zone (× 4)). On the right is the experimental set-up for the transfer tea-making task, annotated with our 6 main AOIs (1 = teabags, 2 = milk, 3 = kettle, 4 = spoon, 5 = mug, 6 = place mat) that were further subdivided into a total of 17 AOIs outlined in Additional file 1

Fig. 4

Performance and gaze data before, during and after training Line plots representing mean (± s.e.m) performance time (a) and performance error (b) in the coin task for both groups across time and the mean (± SD) target locking scores (c) and gaze shifting times (d) at baseline, retention and delated retention specific to the Lift phase of the task

Fig. 5

EEG data before, during and after training Scalp topoplots (top) representing the global distribution of alpha power for each group across the three time points. Displayed below are line plots representing the mean high-alpha inter site clustering (± s.e.m) between T7-Fz (left) and T8-Fz (right) for the MT and GT groups across time

Fig. 6

Relationship between gaze indices and conscious movement control Scatter plots displaying the relationship between TLS and T7-Fz (top row), and between gaze-shifting times and T7-Fz (bottom row), across three time points. Each plot displays the line of best fit (in red) with 95% confidence intervals (shaded in grey), the shared variance (r²) and the significance value (p) of each regression

Fig. 7

EEG data related to the transfer tea-making task Transfer tea-making task data showing scalp topographies representing regional alpha (left), mean (± SD) TLS and gaze shifting times (top-right) and mean (± s.e.m) T7-Fz and T8-Fz EEG connectivity (bottom-right) at both retention and delayed retention for both training groups