Non-Invasive Estimates of Local Field Potentials for Brain-Computer Interfaces: Theoretical Derivation and Comparison with Direct Intracranial Recordings

Abstract

people still believe that only invasive approaches will provide natural and flexible control of robots The rationale is that surgically implanted arrays of electrodes will be required to properly record the brain signals because the non-invasive scalp-recordings with the EEG lack spatial resolution. However, recent advances in EEG analysis techniques have shown that the sources of the electric activity in the brain can be estimated from the surface signals with relatively high spatial accuracy (6-9mm). This resolution compares with the resolution provided by the methods used to detect activation in functional magnetic resonance imaging (fMRI) using 1.5/3 Tesla machines. Note that this is very different to the resolution of the anatomical images provided by the MRI. Aiming at combining the benefits of both approaches, we propose to rely on the non-invasive estimation of local field potentials (eLFP) in the whole human brain from the scalp measured EEG data using a recently developed distributed linear inverse solution termed ELECTRA (Grave de Peralta Menendez et al., 2000). The use of linear inversion procedures yields an on-line implementation of the method, a key aspect for real-time applications. The development of a brain interface based on ELECTRA-i.e., non-invasive estimates of LFP-would allow for methods identical to those used for EEG-based brain interfaces but with the advantage of targeting the activity at specific brain areas. In this respect our approach aims to parallel the invasive approaches described before that directly feeds intracranial signals into the classification stage of the brain interface, except that we calculate these intracranial signals from the surface EEG data. An additional advantage of our approach over scalp EEG is that the latter represents the noisy spatio-temporal overlapping of activity arising from very diverse brain regions; i.e., a single scalp electrode picks up and mixes the temporal activity of myriads of neurons at very different brain areas. Consequently, temporal and spectral features, which are probably specific to different parallel processes arising at different brain areas, are intermixed on the same recording. For example, an electrode placed on the frontal midline picks up and mix activity related to different motor areas known to have different functional roles such as the primary motor cortex, supplementary motor areas, anterior cingulate cortex, and motor cingulate areas. In addition, the proposed approach bears two main advantages over invasive approaches. Firstly, it avoids any ethical concern and the medical risks associated to intracranial electrocorticographic recordings in humans. Secondly, the quality of the signals directly recorded on the brain deteriorates over time requiring new surgical interventions and implants in order to keep the functionality of the device. In this chapter we describe in detail the theoretical framework needed for the non invasive estimation of Local field potentials, the rationale for its application and compare these estimates with the raw EEG (used to estimate the eLFP). To shed some light on the question of the feasibility of non-invasive brain interfaces to reproduce the prediction properties of the invasive systems, we compare the classification results of eLFP non invasively estimated from the EEG with intracranial recordings (IR) during a visuo-motor task.