At neuroTUM, we are dedicated to working on neuroengineering projects that integrate knowledge from both neuro- and computer science to tackle real-world issues. The experimental design in this field varies depending on the project at hand. In this guide, we will outline the step-by-step process of conducting experiments to develop a brain-computer interface (BCI). The experiments we conduct are based on electroencephalography (EEG), a non-invasive technique that records the electrical activity of the brain through sensors placed on the scalp.
It is important to note that different neural activities in the brain correspond to different tasks. For example, attempting to solve a complex problem activates the prefrontal cortex, while executing or imagining movement in the left hand activates the right hemisphere of the primary motor cortex (vice versa for the right hand). EEG can capture these activation patterns, which can then be classified using machine learning algorithms. The experimental design team's objective is to identify a set of actions that can be distinguished from background noise and other actions.
When we founded neuroTUM, our initial set of tasks included cognitive tasks, as well as motor imagery tasks for both the right and left hand, as well as the legs. Our first experiments aimed to compare mental arithmetic (multiplying two numbers) and geometric rotation (imagining rotating an object) to determine the best-suited cognitive task for the BCI. To help describe our experimental design team’s work in the following sections, we will introduce this previously mentioned example of finding the optimal cognitive task.
To start, the experiment's objective must be clearly defined. In our case, we aim to compare the distinguishability between mental arithmetic and geometric rotation tasks from motor imagery tasks.
After defining the objective, the testing environment must be considered. In our case, we will use a black screen that displays a set of different symbols (e.g arrows, capital letters) each representing the task the pilot should perform. To prevent the subjects from getting bored, the symbols are randomly shuffled. The timings for the task are based on existing literature and adjusted based on the subject's preferences.
To minimize the impact of visual evoked potentials* on the EEG data, we begin the task only after the visual cue has disappeared. This is an essential step to ensure the EEG signal obtained reflects the neural activity associated with the task.
The final step in experimental design is configuring the electrode layout around the regions of interest. This is crucial as the EEG signal is recorded from specific areas of the scalp that correspond to different brain regions involved in the task. By placing the electrodes in optimal positions, we can obtain a better signal-to-noise ratio and improve the accuracy of the data obtained.
Have you ever tried to imagine a movement without actually performing it? Brain-Computer-Interfaces rely on these counterintuitive tasks to develop a technology applicable for paralyzed patients. It is, however, quite difficult to perform motor imagery tasks. To ensure a strong and reliable signal, our subjects practice the experiment at home for at least 30 minutes without wearing an EEG hat.
For the experiment, the electrodes have to be applied on the planned positions. The impedances between the scalp and the electrodes are then reduced by applying conductive gel. Finally, once the pilot is seated in a defined position to improve reproducibility, the experiment is performed – as long as the pilot feels fit of course (usually about 200 repetitions of the above mentioned steps).
Throughout the experiment, the code responsible for the virtual environment continuously provides updates on the progress of the experiment, such as which task is being performed and if there are any breaks. This data stream is synchronized with the EEG data and recorded in real time.
Would you like to perform an EEG experiment? Did you find flaws in the proposed experiment or want to propose a better one? Let us know about your ideas and help us develop the most reliable brain-computer-interface!
*Visual evoked potentials are the neural response to visual stimuli. They can also be used for BCI applications, however as we want our model to learn only the neural response to the corresponding task not our specific test environment we are not using them.
Article written by Leon
Edited by Charlie, Fatma and Isabel