Biomedical Engineering Reference
In-Depth Information
transduction from the sensory organs (hair cells in hearing or photoreceptor cells in vision) or to the
muscles (spinal cord injury or peripheral neuron damage) is broken, the following neural engineering
question is appropriate: “What is the best way to deliver stimuli to the sensory cortex or to extract
information about voluntary movement from motor cortex activity?” Obviously, there are two parts
to the answer: one relates to the actual design and implementation of the engineering systems that
can translate sound or light into electrical stimuli, control motor neurons in the periphery, or directly
activate a prosthetic and the other relates to the neurophysiology knowledge to help prescribe the
spatiotemporal characteristics of these stimuli to mimic the ones created by the sensory pathway to
the brain or, in the case of the motor BMIs, to help model the spatiotemporal interactions of neural
activity in the motor cortex. The disciplines of engineering and neuroscience (at least) are involved,
and none of them independently will be able to solve the problem. Therefore, this is an imminently
multidisciplinary enterprise that requires large research groups to be productive.
1.2 BEyoNd STaTE-oF-ThE-aRT TEChNology
One may think that the engineering side is better prepared to deliver systems in the short term be-
cause of the computational speed of modern digital signal processors (DSPs) and technological in-
novation in sensors, miniaturization, and computers, but this is illusory; interfacing with the central
or peripheral nervous systems required beyond state-of-the-art technology and modeling on many
fronts. Engineers working in BMI rapidly develop an appreciation for the solutions encapsulated
in biological systems because of the demanding specifications of the systems. To relatively assess
the relationships between the performance specifications, let us look at a space with three axes (as
shown in Figure
1.2
) that relates scale, power dissipation, and computation needed to interface
brain tissue and microelectromechanical computer technology.
From the figure, we can see that the relationship of the space spanned by the current technol-
ogy is very different from that of the BMI system, and only a small portion of the spaces overlap
(purple). If we compare the scale of our current multiple-input-multiple-output (MIMO) systems,
there is a large disparity because the brain is a huge computational system, with, on the order of
10
12
neurons, about 10
15
synapses [
10
]. The density of neurons in the neocortex of a human is about
10
4
/mm
3
but can vary as a function of the type of animal and the spatial location (i.e., visual, so-
matosensory, etc.). The conventional electrophysiological techniques (discussed later) present only
a coarsely subsampling the neuronal population. The electronics needed to acquire the activity of
this small sampling are very sophisticated. First, the amplifiers have to provide high common-mode
rejection ratio, high gains with very small power. Second, the system has to transmit very large data
rates also with constraints in power. To make the transmission more reliable, the analog signal must
be converted to a digital representation, which adds substantially to the total power consumed.
