Introduction to Neural Interfaces - Brain-Machine Interface Engineering - page 8

1.4 gENERaTIoN oF CoMMUNICaTIoN aNd CoNTRol
SIgNalS IN ThE BRaIN
The generators of the “landscape of basins and attractors,” as described by Freeman, are the electro-
physiological modulation of neural networks in the living tissue. From a methodological perspective,
access to the spatial scale has played an important role in BMI development and includes the use of
electrophysiological methods to extract spikes (action potentials) [ 33 ], local field potentials (LFPs)
[ 34 ], electrocorticogram (ECoG) [ 35 ], and EEG [ 36 ]. 4 Figure 1.4 shows the relative relationships
of the spatial resolution for the three most common electrophysiological techniques used for BMIs.
At the highest resolution, microwire electrode arrays can sense the activity of single neurons simul-
taneously; however, they are the most invasive to the tissue. ECoG recording is less invasive because
the electrodes rest directly on the cortex, and EEG electrodes are the least invasive because they sit
on the scalp. Deriving signatures of sensation or intent from electrophysiological recordings carries
either the constraints of model selection or the uncertain hypotheses of the signal generation at each
level of scale. For instance, microelectrode arrays that sense extracellular action potentials are very
specific to neuronal function but are local in space and carry the bottleneck of sensing a representa-
tive ensemble of neurons within the technical limitations of electrode spacing and number and the
difficulty of bridging the time scale to cognitive events. Because BMI paradigms have developed,
the viewpoint from the single neuron has expanded to encompass the time-dependent communica-
tion of ensembles of hundreds of neurons [ 33 ]. These principles have been directly applied with
great success to BMI architectures [ 37-40 ] that utilize only a single-unit activity (SUA). Recently,
the use of LFPs have also been shown to improve BMI performance when combined with an SUA
[ 34 ]. Therefore, harnessing both the inputs (dendritic activity) and outputs (action potentials) of
neural assemblies seems to be critical for interpreting the intent of the individual. 5 Using a much
larger scale, scalp EEG contains a summation of dendritic and axonal currents of millions of neu-
rons over the cortical impedance. The continuous amplitude macroscopic signals modulate at the
time scale of behavior but lack specificity and can be much more influenced by noise. The ECoG,
when compared with the EEG, has the great advantage of being more spatially specific (smaller
electrodes), much less contaminated by noise, and can display better frequency resolution because
of its proximity to the neural tissue.
4 Electrophysiological recordings are one particular subset of brain sensing methodologies. Other techniques such
as the magnetoencephalogram, near-infrared spectroscopy, and functional magnetic resonance imaging have been
used in BMI studies. However, the aforementioned techniques require quite specialized hardware and provide much
different spatial and temporal resolutions compared with the electrophysiological techniques.
5 In the next chapter, we will present a brief review of the generation of electrical potentials in excitable neural tissue.