A neural network based cybernetic interface for identification of simple stimuli based on electrical activity in an intact whole nerve

The cybernetic interface design explored in this research utilized a cuff electrode as the detector. The cuff electrode was placed around the intact, ensheathed lateral eye optic nerve of the horseshoe crab Limulus polyphemus and recorded the bulk activity during presentation of visual stimuli to the lateral eye. The cuff was formed from a short length of silicone tubing and contained multiple differential pairs of electrodes placed around the lumen. By virtue of their varied locations each electrode pair recorded a different admixture of the sources which comprised the bulk activity. The recorded data was filtered using wavelet based signal processing. Two information theoretic measures, the j-divergence and entropy, were computed on the de-noised data. The j-divergence was used to assess the overall quality of the data and predict the system performance during pattern classification. The entropy was used to build training/testing sets for a neural network. Classification of entropy data into the relative probability of one or more of the stimulus locations being illuminated was accomplished using a cascade correlation neural network.; The system was able to distinguish between activity due to stimulus and that arising from noise/spontaneous firing with 100% accuracy. The system's accuracy at determining which location(s) on the lateral eye were stimulated based on the electrical activity in the optic nerve was also very good.; The system discussed in this thesis is a significant step toward the realization of a minimally invasive neural interface. Such an interface could be used in a practical setting, such as controlling a prosthetic limb or remotely operated robotics; or it could be used as a creative device to allow natural interaction in a computer generated environment. The potential for such an interface is vast and could prove to be an important step in our future.