Authors:
M. Tangermann
1
;
J. Reis
2
and
A. Meinel
1
Affiliations:
1
Brain State Decoding Lab, Cluster of Excellence BrainLinks-BrainTools and Albert-Ludwigs-University, Germany
;
2
University Medical Center Freiburg, Germany
Keyword(s):
EEG, Oscillatory Components, Single-Trial Analysis, Performance Prediction, Machine Learning, Spatial Filters, Subspace Decomposition, Hand Motor Task, Isometric Force, Performance Metrics, Reaction Time, Jerk, Motor Rehabilitation.
Related
Ontology
Subjects/Areas/Topics:
Biomedical Engineering
;
Biomedical Instruments and Devices
;
Brain-Computer Interfaces
;
Devices
;
EMG Signal Processing and Applications
;
Health Engineering and Technology Applications
;
Human-Computer Interaction
;
Neural Rehabilitation
;
Neural Signal Processing
;
NeuroSensing and Diagnosis
;
Neurotechnology, Electronics and Informatics
;
Outcome Measures
;
Physiological Computing Systems
Abstract:
The power of oscillatory components of the electroencephalogram (EEG) can be predictive for the single-trial performance score of an upcoming task. State-of-the-art machine learning methods allow to extract such predictive subspace components even from noisy multichannel EEG recordings. In the context of an isometric hand motor rehabilitation task, we analyse EEG data of n=20 normally aged subjects. Predictive oscillatory EEG subspaces were derived with a spatial filtering method (source power comodulation, SPoC), and the transfer of these subspaces between five performance metrics but within data of single subjects was investigated. Findings suggest, that on the grand average of 20 subjects, informative SPoC subspace components were extracted, which could be shared between a set of three metrics describing the duration of subtasks and jerk characteristics of the force trajectories. Transfer to any other of the remaining four metrics was not possible above chance level for a metric d
escribing the reaction time and a metric assessing the length of the force trajectory. Furthermore we show, that these transfer results are in line with the structure of cross-correlations between the performance metrics.
(More)