A new descriptor of neuroelectrical activity during BCI–assisted Motor Imagery training in stroke patients Manuela Petti 1,2 , Donatella Mattia 1 , Floriana Pichiorri 1 , Laura Astolfi 1,2 , Febo Cincotti 1,2 1 IRCCS Fondazione Santa Lucia, Rome, Italy. 2 Dept. of Computer, Control, and Management Engineering, Univ. of Rome “Sapienza”, Italy. manuela.petti@uniroma1.it Abstract Recent BCI applications stroke motor rehabilitation have raised important concerns regarding the type of brain activity which one would train in agreement with an evidence-based approach in rehabilitation. In this pilot study we proposed an offline analysis on EEG data acquired during a BCI-assisted motor imagery training performed by a stroke patient, with the aim of defining an index for the evaluation of the training achievements across session. The proposed h parameter would be independent from the selected BCI training setting and would better describe the physiological properties of the patterns generated during training, allowing a more appropriate evaluation of the training achievements than the behavioral performance (i.e. percentage of hit target). 1 Introduction Nowadays, Brain Computer Interface (BCI) represents a promising technology to support motor and cognitive rehabilitation after stroke. In such rehabilitative context, BCI application aims at increasing the neuroelectric or metabolic brain responsiveness, which in turn would lead to a better recovery of function. The Electroencephalographic (EEG) -based BCI operated by motor imagery (MI) can provide a valuable approach to support mental motor practice to enhance arm motor recovery after stroke [Mattia et al., 2012]. However, as stroke cortical lesions may result in a functional reduction/derangement of neuroelectrical activity generated over the ipsilesional hemisphere there is a need for further implementation of the procedures for recognition of those EEG patterns which are reinforced during the BCI-supported training of MI. Furthermore the online classification of trials as successful and failed also relies on a arbitrary choice of parameters and gains that do not strictly reflect the intrinsic properties (i.e. the level of desynchronization of SMR) of the patterns of activity recorded during the training. The aim of this study was to define an index which would be independent from the settings adopted for the online control and thus, would describe the properties of neuroelectrical activations across BCI training sessions more appropriately than the hit rate (behavioral performance). To this purpose, we performed an offline analysis of EEG data sets acquired from stroke patients who underwent a MI- assisted BCI training aiming at promoting functional motor recovery of the paralized upper limb [Pichiorri et al., 2011]. The estimated index was monitored across training sessions and used to sort trials according to their intrinsic properties. Proceedings of the 6th International Brain-Computer Interface Conference 2014 DOI:10.3217/978-3-85125-378-8-94 Published by Graz University of Technology Publishing House Article ID 094-1