Neuralprosthetic Devices: Inputs and Outputs.

Abstract

Prior studies have demonstrated that the firing rate of cortical neurons can be volitionally modulated by a subject to generate a controllable output signal; this neural output signal can then be manipulated to direct a robotic arm, a cursor on a computer screen, or other interface device. The burgeoning field of neural control has led to a number of innovative applications, known more commonly as neuroprosthetic devices. Neuroprosthetic devices have the potential to return some degree of functionality to the over 250,000 Americans with incapacitating spinal cord injuries, or allow healthy subjects to control electronic devices in their everyday lives. The research presented here consists of three studies focused on improving the current generation of neuroprosthetic devices.

In the first study, we introduced and evaluated a Bayesian maximum-likelihood estimation (bMLE) strategy to identify optimized training data for neuroprosthetic devices. By limiting initial decoding assumptions and training only on relevant neural data, accurate neural-control was possible with as few as two neurons, using minimal training data and no a-priori¬ movement measurements for calibration. Moreover, implanted subjects obtained useful prosthetic control using local field potentials and neurons from cingulate cortex as input.

In the second study, we refined a method to electrochemically deposit surfactant-templated ordered poly(3,4-ethylenedioxythiophene) (PEDOT) films on the recording sites of standard “Michigan” probes, and evaluated the in vivo efficacy of these modified sites in recording chronic neural activity. PEDOT sites were found to outperform control sites in terms of signal-to-noise ratio and number of viable unit potentials - thereby improving the quality of neural input sources to the neuroprosthetic device.

In the third study, we evaluated a technique known as common average referencing (CAR) to generate a more ideal reference electrode for microelectrode recordings. CAR was found to drastically outperform standard types of electrical referencing, reducing noise by more than 30 percent. As a result of the reduced noise floor, arrays referenced to a CAR yielded almost 60 percent more discernible neural units than traditional methods of electrical referencing – again improving the quality of neural input sources to a neuroprosthetic device.