Hand motor performance decoding using slow potentials
Simple repetitive hand motor tasks, such as those used for rehabilitation training after stroke, are known to show strong trial-by-trial performance variations even though a trial typically lasts a few seconds only. This performance variation are observed not only in patients, but also for healthy users.
We assume, that the brain state of the user prior to the task execution plays an important role for the upcoming motor performance. We found, that neural oscillations measured by, e.g., the electroencephalogram (EEG) can provide information about the expected performance of the upcoming trial a few hundred milliseconds prior to the start of the trial.
In this project, it shall be investigated, if information about the upcoming performance can be gained also from non-oscillatory components of the EEG signal. Specifically, the slow Bereitschaftspotential / readiness potential may also be present prior to the start of each trial and may be informative to predict the hand motor performance. If time allows, the search can be expanded to other signal components by using convolutional neural networks and a post-hoc analysis of features learned.
The candidate will perform an analysis of existing data from healthy users and stroke patients in order to investigate the research question. The candidate can built upon existing data analysis pipelines in Python and Matlab.
- Machine learning
- Good programming skills in Python (familiarity with numpy, sklearn, at least one popular deep learning framework).
- Good mathematical background and intuition.