The issue of benchmarking the performance of various neural recording probes is a difficult problem.  A number of papers have been published or been submitted evaluating the long term performance of various microwire assemblies, the Utah array, and Michigan probe technology (Williams, JC 1999; Nicolelis, M. A. L. 2003; Vetter 2004; Selim Suner 2005). In 1999, Justin Williams, then a graduate student Dr. Kipke’s group, reported that microelectrode arrays based on microwires implanted in cerebral cortex of eight subjects yielded discriminable unit activity on 80 percent of the electrodes; one subject exhibited unit activity on 94 percent of recorded electrode sites as far as 226 days post surgery (Williams, JC 1999). Also using a high density microwire array, Prof. Nicolelis at Duke University implanted three monkeys with from 96 -704 microwires per subject in as many as 10 cortical locations (Nicolelis, M. A. L. 2003). Fifty-four percent of the implanted microwires yielded at least one recordable cell three to four weeks after the surgery, with an average signal to noise ratio of better than five to one (Nicolelis, M. A. L. 2003). One subject demonstrated unit activity more than eighteen months after implantation (Nicolelis, M. A. L. 2003). More recently, Rio Vetter implanted ten rodents with a total of fourteen Michigan silicon-substrate microelectrode arrays, noting that more than 90 percent the probe sites consistently recorded discrimnable activity out as far as 127 days with an average signal to noise ratio of greater than seven to one (Vetter 2004). As of this writing, Selim Suner from Prof. Donoghue’s group at Brown University has submitted a paper to IEEE detailing the results from three Macaque monkeys implanted with Bionic (Cyberkinetics, Inc) silicon probe arrays (Selim Suner 2005). Suner notes that eighty percent of the electrode sites recorded high quality neural waveforms with an average signal to noise of 4.8 to one (Selim Suner 2005). Waveforms were evident as long as 569 days after implantation (Selim Suner 2005). The above list is not exhaustive, but is indicative of the types of studies that have been conducted to date exploring the reliability of available options for chronic recording implants. All of these studies use subjective metrics to quantify performance over a limited number of subjects. In all four cases, single units were identified by various visual sorting techniques, which have been shown to be variable (Wood 2004). As a result, these studies are not directly comparable nor do they have the statistical power to adequately characterize the devices in question. Moreover, these studies do not adequately address the stability of individual units over time, a requirement of a number of neuroprosthetic and neural research thrusts.

The relatively number of prospective CNCT participants, coupled with the combined expertise available through the Center, gives us the unique opportunity to generate a new set of objective metrics to quantify recording performance and widely apply these metrics in a number of research settings in order to generate a more comprehensive data set. This comprehensive data set will be used to establish baseline levels of performance for neural recordings that have the statistical power for comparative studies. The Kipke lab at Michigan has developed an automated procedure for objectively sorting unit potentials from high speed recordings, based heavily on earlier work in this area (Lewicki 1998; Zouridakis 2000). Moreover, the CNCT has developed standardized techniques for obtaining high speed recordings, impedance spectroscopy, and histological analysis. Through our training program, we will educate end users in how to consistently apply these techniques. With the help of our vast community of probe users, we have the opportunity to generate an unparalleled data set suitable for powerful, rigorous statistical analysis. It is our hope to establish an objective methodology for probe evaluation that will be adopted by the rest of the neural systems community.