Motor Imagery Neurofeedback Reveals Performance Differences in Single vs. Competitive Task

Motor imagery neurofeedback is a powerful tool for training the brain and supporting neurorehabilitation. but its success depends on many factors. While much work has focused on signal processing and user traits, the role of social context is less explored.

This study asks a simple but important question: does performing neurofeedback alone differ from doing it in a competitive setting?

Setting up competitive Motor Imagery neurofeedback training

During the experiment, the participant had a task to move a NAO humanoid robot by imagining the movement.

To read the activation of the sensorimotor cortex during MI, authors used real-time EEG data. They used mbt’s portable Smarting device to measure it. Before we go into details on how they decoded EEG data in real-time, let’s first look at the experimental blocks.

The end-goal of the task was to get the participants to move the robot as far as they could with MI. The experiment had 4 blocks. The first one was just them mimicking the walking i.e. physically doing the movement.

The seconds block they had to imagine the movement, but there was no feedback from the robot. So, this is an MI but no FB block. EEG data collected during this block is used for parameter estimation that will be used for the FB in the next blocks.

The last two blocks (3^rd and 4^th) consisted of both MI and FB from the humanoid robot.

For all 25 participants the experiment is repeated twice: single- and multi-user condition. For the competition, the opponent also wore an EEG headset but with no real motor imagery neurofeedback.

In movement-related paradigms, event-related desynchronization (ERD) plays a key role.

ERD refers to power changes in rhythmic brain activity recorded over sensorimotor areas. These changes occur within the mu (8–12 Hz) and beta (13–30 Hz) frequency ranges.

In this study, researchers calculated relative ERD for each EEG channel. Relative ERD is ERD during the task in relation to the ERD in the baseline before the task. This step is done after the 2^nd experimental block. Then, the channel with the strongest relative ERD is used for the MI + FB blocks (3^rd and 4^th).

During the two last blocks, authors applied linear discriminant analysis (LDA) on real-time EEG data. LDA is a simple machine-learning classifier that tries to separate two or more classes by finding a linear boundary.

From the training set, they looked at the distribution of the classifier’s output values. They split this distribution into quartiles (lower, median, upper). Depending on this output, the robot would eventually move: 1, 2 or 3 steps!

No significant difference in ERD between single and competitive conditions

This does match prior findings, that say social context doesn’t always result in better performance. In this case, single and competitive conditions didn’t cause a different brain response, therefore, performance.

Since they found no significant difference, researchers divided the participants for an additional interindividual analysis.

Image above shows a division of participants in 2 groups. Authors separated them into single-gain (performed better single) and competitive-gain (performed better in competitive setup).

Single-gain group: Showed significantly stronger ERD in the single vs. training condition. Suggests they benefit from the single condition.

Competitive-gain group: Didn’t show significant gains compared to training.

So social context seems to matter differently across individuals. For some, competition might be distracting, maybe due to added cognitive load or attention difficulties.

Additionally, both groups show weaker ERD during training. Social context can elevate ERD above baseline. But this effect isn’t uniform across users in this study. More research is needed to confirm whether this pattern is systematic.

This finding supports a move toward user-centered designs in motor imagery neurofeedback systems.

Participant’s preferred condition didn’t align with their MI neurofeedback performance

During the experiment participants gave subjective ratings:

1. Easiness of MI
2. Vividness of MI
3. General performance motivation
4. General control beliefs in dealing with technology

Participants reported being motivated and MI experience to be easy and vivid, on average.

Based on self-reports, participants’ preferred condition didn’t align with the ERD-based performance. Some other, unidentified factors could be looked at as potential predictors here – such as motor skills, visual processing, gaming experience etc.

Prediction of stronger relative ERD?

The researchers wanted to find out if it’s possible to predict which social context works best for each participant. In other words, could they know in advance if someone would perform better in the single or competitive task.

They found that the ERD measured during the training block predicted which social context would lead to stronger ERD. The subjective scores didn’t add any predictive value.

Brain activity during training could then be used to predict individual optimal conditions. Because training data is already collected in most neurofeedback setups, this means the best social context for each participant could be predicted with little extra effort.

 

This study shows that social context can shape motor imagery neurofeedback performance, but the effect isn’t the same for everyone. Some participants performed better alone, while others benefited more from competition. Crucially, brain activity during training was the best predictor of which setup worked for each person. These results highlight the importance of tailoring neurofeedback to individual users and point toward more personalized approaches in the future.

You can read more about the results in the original paper by Mareike Daeglau et al.