Introduction to Neural Interfaces - Brain-Machine Interface Engineering - page 6

such that proper signal processing models can be designed and applied. The neural electrode inter-
face is of fundamental importance because invasive BMIs require prolonged contact with the neural
tissue and the electrical characteristics of the interface should be stable and reliable to guarantee
long life, high signal-to-noise ratios (SNRs), and avoid cell death. By far, the most compelling
explanation for signal degradation in chronic neural recordings is the series of events that occurs
after implantation, which include inflammatory response, a disruption of the blood-brain barrier,
and initiation of astrogliosis and recruitment of microglia and macrophages to the insertion site
[ 13-16 ]. Each of these problems has been studied extensively, and a variety of approaches have
been proposed to improve the electrode tissue interface, which include novel electrode material
coatings (conductive polymers, ceramic matrices, timed release of dexamethasone) and the use of
microfluidics for engineering the glial response and tissue inflammation [ 5 , 17 ]. Several groups have
demonstrated that it is possible to collect data from neurons for months, but there is a long road still
ahead to control tissue responses to the implant.
1.3 CoMPUTaTIoNal ModElINg
The main problem that directly interests the readers of this topic is the issue of modeling neural inter-
actions, which will be discussed extensively throughout the text. At this point, we would like to only
summarize the state of the art in computational neuroscience and a broad-stroke picture of the signals
used in BMIs. There is extensive neuroanatomical knowledge on the excitability properties of neurons
as well as the functional mapping of the interconnection among cortical and subcortical networks
[ 18-20 ]. The issue is that the interaction principles among neurons, which implement the so-called
neural code [ 21 ], are poorly understood. Moreover, large pieces of a theoretical framework that defines
neural representation (i.e., mechanisms of stimulus activated elements), computation (i.e., how the ele-
ments optimally use the stimulus), and dynamics (i.e., how the elements change in time) are missing.
From a pragmatic point of view, the fundamental constructs of the brain (or any biological
system for this matter) do not come with an instruction manual. The task of finding out how the
system is working and what is being computed has been a significant difficulty for computational
neuroscientists. Compared with most mechanical or electrical systems, the problem is perhaps even
harder because of the self-organizing bottom-up principles embedded in biology versus the top-
down approach used in engineering design. The difference in the two perspectives, as shown in
Figure 1.3 , is the necessary adoption of a big-picture approach by computer scientists and a re-
ductionist approach by many neuroscientists. The top-down approach arises from an engineer-
ing perspective where the broad goals are to “design a machine to perform a particular task.” The
bottom-up perspective is one of the phenomenologist or the taxonomist: “collect data and organize