A parameter free model for motor imagery detection based on Riemannian geometry: preliminary results

Abstract

Brain-computer interfaces (BCIs) provide a non-muscular channel to control external devices using only brain activity. Motor Imagery BCI (MI-BCI) systems are based on decoding the imagination of certain movements. Although the Common Spatial Patterns (CSP) algorithm, as well as its regularized versions, can be successfully applied for MIdetection, it has some limitations in adapting to data changes. In this context, Riemannian geometry seems to be a promising approach to construct a simple, robust and parameter-free decoding model. In this work we implement and evaluate an MI decoding algorithm based on Riemannian Geometry. In particular, the Riemannian distance and its mean are used for constructing a "minimum distance to mean" (MDM) classifier. MDM is compared with the traditional CSP method, showing very similar classification results in both cross-validation and online simulation scenarios. These results indicate that Riemannian framework seems to be a very promising tool for robust MI detection.