Software Requirements

The second significant issue is that of the computational requirements that would allow the reading, storing and contextualizing of the enormous amount of neuronal information that would become available with the vascular approach described above. While this may prove to be more challenging than the hardware component of this interface, it would also be most valuable, as the proper understanding of such activity would give us an significant window into brain function, further defining the relations between electrophysiology and cognitive/motor properties of the brain.

Attempting to investigate this problem, the second step in this proposal, would be the development of mathematical algorithms able to classify brain states based on neuronal unit activity and field potential analysis. Initially, we plan to correlate, in real time, the moment-to-moment electrical activity of neurons with large functional brain states. It is assumed that the electrical properties of neurons define all possible brain states and that such states co-vary systematically with the global state dynamics. However, this does not imply that there exists one-to-one correspondence between purely local patterns of brain activity and a particular set of functional states. The generation of a new functional state in the brain, for instance, transition "sleep-wakefulness," is known to correspond to activity reorganization over many groups of neurons. Needless to say, there is a large number of possible patterns that differs minimally from one other. The approach is to map the small variance patterns into relatively small sets of different functional states. For example, in the simplest case only three global functional states may be considered: (1) sleep, (2) wakefulness, and (3) "none of the above" or uncertain state, e.g., drowsiness. The last state is an absolutely necessary form to be included, for two reasons: (a) mathematically, the output domain of the algorithm must be closed in order to address correctly "any possible input pattern," including those that have unavoidable noise impact or belong to intermediate, non-pure states without a reliable answer within statistical significance level; and (b) from the conceptual viewpoint, the third state is vital, as for instance, seeing can only occur during wakefulness, and during sleep, this state is uncertain.

The design of the hardware part of the interface (see Figure C.10B) has not been dictated by electronic purposes only but also pursues the goal of preliminary signal processing. Here, we use the commonly accepted hypothesis that neurons interact with each other mostly via action potentials and related synaptic interactions. Thus, it seems to be natural to convert electrical signals taken from n-electrodes into binary form. This approach has many advantages. In particular, if the threshold level for digitalization is appropriately chosen, we would be able to overcome the following problems:

• Not all electrodes would be placed at "right" positions (some of them may be far enough from any neuron to produce reliable data), or just damaged.

• Two electrodes placed in vicinity of a single neuron but at diverse distances from it will produce output voltage traces of different amplitude.

• The signal-to-noise ratio may not be optimal if an electrode records from more than one neuron, as one of them may be selected and others suppressed by the threshold system.

Moreover, binary form is computer friendly and supports efficient operation. Also additional processing logic can be easily included between a computer and the terminals of microwires that would significantly speed up data acquisition, storage, and contextualization.

0 0

Post a comment