A study on the efficient tracking of ERD based on the adaptive identification of the subject's reactive band

Imagining a certain movement produces changes in the subject's EEG motor rhythm in the same manner in which the real execution of the movement does. These variations are reflected by Even-Related Desynchronizations (ERD) and appear asa magnitude decrease of the frequency components comprising the mu band. By monitoring the amplitude and power changes, the automatic detection of imagery or movement can be conducted. Furthermore, features can be extracted and used for real-time classification of these events. It is known that in most cases, the majority of changes are constrained within a narrow frequency interval representing the subject's reactive band. This provides means to increase the efficiency of interpreting desynchronizations by working only with the most representative information. The paper presents three improved methods of adaptively modelling the EEG signal, based on a Band Limited Multiple Fourier Linear Combiner (BMFLC). The time-frequency decomposition thus obtained is subsequently exploited in order to automatically identify the subject's specific reactive band. This process is conducted by detecting the highest decrease in power with the purpose of using a selection of power information as features for the classifier. The final goal of the investigated algorithms is to allow discrimination between left and right hand motor imagery, as demanded by a brain computer interface rehabilitation system.