Justin C. Sanchez

Comprehensive characterization and failure modes of tungsten microwire arrays in chronic neural implants

For nearly 55 years, tungsten microwires have been widely used in neurophysiological experiments in animal models to chronically record neuronal activity. While tungsten microwires initially provide stable recordings, their inability to reliably record high-quality neural signals for tens of years has limited their efficacy for neuroprosthetic applications in humans. Comprehensive understanding of the mechanisms of electrode performance and failure is necessary for developing next generation neural interfaces for humans. In this study, we evaluated the abiotic (electrophysiology, impedance, electrode morphology) and biotic (microglial reactivity, blood-brain barrier disruption, biochemical markers of axonal injury) effects of 16-channel, 50 µm diameter, polyimide insulated tungsten microwires array for implant durations that ranged from acute to up to 9 months in 25 rats. Daily electrode impedance spectroscopy, electrophysiological recordings, blood and cerebrospinal fluid (CSF) withdrawals, and histopathological analysis were performed to study the time-varying effects of chronic electrode implantation. Structural changes at the electrode recording site were observed as early as within 2-3 h of electrode insertion. Abiotic analysis indicated the first 2-3 weeks following surgery was the most dynamic period in the chronic electrode lifetime as there were greater variations in the electrode impedance, functional electrode performance, and the structural changes occurring at the electrode recording tips. Electrode recording site deterioration continued for the long-term chronic animals as insulation damage occurred and recording surface became more recessed over time. In general, electrode impedance and functional performance had smaller daily variations combined with reduced electrode recording site changes during the chronic phase. Histopathological studies were focused largely on characterizing microglial cell responses to electrode implantation. We found that activated microglia were present near the electrode tracks in all non-acute animals studied, thus indicating presence of a neuroinflammatory response regardless of post-implantation survival times and electrode performance. Conversely, dystrophic microglia detectable as fragmented cells were found almost exclusively in acute animals surviving only few hours after implantation. While there was no consistent relationship between microglial cell responses and electrode performance, we noticed co-occurrence of high ferritin expression, intraparenchymal bleeding, and microglial degeneration suggesting presence of excessive oxidative stress via Fenton chemistry. Biochemical analysis indicated that these electrodes always caused a persistent release of axonal injury biomarkers even several months after implantation suggesting persistent tissue damage. Our study suggests that mechanisms of electrode failure are multi-factorial involving both abiotic and biotic parameters. Since these failure modes occur concurrently and cannot be isolated from one another, the lack of consistent relationship between electrode performance and microglial responses in our results suggest that one or more of the abiotic factors were equally responsible for degradation in electrode performance over long periods of time.

A comparison of optimal MIMO linear and nonlinear models for brain–machine interfaces

The field of brain-machine interfaces requires the estimation of a mapping from spike trains collected in motor cortex areas to the hand kinematics of the behaving animal. This paper presents a systematic investigation of several linear (Wiener filter, LMS adaptive filters, gamma filter, subspace Wiener filters) and nonlinear models (time-delay neural network and local linear switching models) applied to datasets from two experiments in monkeys performing motor tasks (reaching for food and target hitting). Ensembles of 100-200 cortical neurons were simultaneously recorded in these experiments, and even larger neuronal samples are anticipated in the future. Due to the large size of the models (thousands of parameters), the major issue studied was the generalization performance. Every parameter of the models (not only the weights) was selected optimally using signal processing and machine learning techniques. The models were also compared statistically with respect to the Wiener filter as the baseline. Each of the optimization procedures produced improvements over that baseline for either one of the two datasets or both.

Coadaptive Brain–Machine Interface via Reinforcement Learning

This paper introduces and demonstrates a novel brain-machine interface (BMI) architecture based on the concepts of reinforcement learning (RL), coadaptation, and shaping. RL allows the BMI control algorithm to learn to complete tasks from interactions with the environment, rather than an explicit training signal. Coadaption enables continuous, synergistic adaptation between the BMI control algorithm and BMI user working in changing environments. Shaping is designed to reduce the learning curve for BMI users attempting to control a prosthetic. Here, we present the theory and in vivo experimental paradigm to illustrate how this BMI learns to complete a reaching task using a prosthetic arm in a 3-D workspace based on the users neuronal activity. This semisupervised learning framework does not require user movements. We quantify BMI performance in closed-loop brain control over six to ten days for three rats as a function of increasing task difficulty. All three subjects coadapted with their BMI control algorithms to control the prosthetic significantly above chance at each level of difficulty.

Ascertaining the importance of neurons to develop better brain-machine interfaces

In the design of brain-machine interface (BMI) algorithms, the activity of hundreds of chronically recorded neurons is used to reconstruct a variety of kinematic variables. A significant problem introduced with the use of neural ensemble inputs for model building is the explosion in the number of free parameters. Large models not only affect model generalization but also put a computational burden on computing an optimal solution especially when the goal is to implement the BMI in low-power, portable hardware. In this paper, three methods are presented to quantitatively rate the importance of neurons in neural to motor mapping, using single neuron correlation analysis, sensitivity analysis through a vector linear model, and a model-independent cellular directional tuning analysis for comparisons purpose. Although, the rankings are not identical, up to sixty percent of the top 10 ranking cells were in common. This set can then be used to determine a reduced-order model whose performance is similar to that of the ensemble. It is further shown that by pruning the initial ensemble neural input with the ranked importance of cells, a reduced sets of cells (between 40 and 80, depending upon the methods) can be found that exceed the BMI performance levels of the full ensemble.

Extraction and localization of mesoscopic motor control signals for human ECoG neuroprosthetics.

Electrocorticogram (ECoG) recordings for neuroprosthetics provide a mesoscopic level of abstraction of brain function between microwire single neuron recordings and the electroencephalogram (EEG). Single-trial ECoG neural interfaces require appropriate feature extraction and signal processing methods to identify and model in real-time signatures of motor events in spontaneous brain activity. Here, we develop the clinical experimental paradigm and analysis tools to record broadband (1Hz to 6kHz) ECoG from patients participating in a reaching and pointing task. Motivated by the significant role of amplitude modulated rate coding in extracellular spike based brain-machine interfaces (BMIs), we develop methods to quantify spatio-temporal intermittent increased ECoG voltages to determine if they provide viable control inputs for ECoG neural interfaces. This study seeks to explore preprocessing modalities that emphasize amplitude modulation across frequencies and channels in the ECoG above the level of noisy background fluctuations in order to derive the commands for complex, continuous control tasks. Preliminary experiments show that it is possible to derive online predictive models and spatially localize the generation of commands in the cortex for motor tasks using amplitude modulated ECoG.

A Trial of Scheduled Deep Brain Stimulation for Tourette Syndrome Moving Away From Continuous Deep Brain Stimulation Paradigms

OBJECTIVE To collect the information necessary to design the methods and outcome variables for a larger trial of scheduled deep brain stimulation (DBS) for Tourette syndrome. DESIGN We performed a small National Institutes of Health-sponsored clinical trials planning study of the safety and preliminary efficacy of implanted DBS in the bilateral centromedian thalamic region. The study used a cranially contained constant-current device and a scheduled, rather than the classic continuous, DBS paradigm. Baseline vs 6-month outcomes were collected and analyzed. In addition, we compared acute scheduled vs acute continuous vs off DBS. SETTING A university movement disorders center. PATIENTS Five patients with implanted DBS. MAIN OUTCOME MEASURE A 50% improvement in the Yale Global Tic Severity Scale (YGTSS) total score. RESULTS Participating subjects had a mean age of 34.4 (range, 28-39) years and a mean disease duration of 28.8 years. No significant adverse events or hardware-related issues occurred. Baseline vs 6-month data revealed that reductions in the YGTSS total score did not achieve the prestudy criterion of a 50% improvement in the YGTSS total score on scheduled stimulation settings. However, statistically significant improvements were observed in the YGTSS total score (mean [SD] change, -17.8 [9.4]; P=.01), impairment score (-11.3 [5.0]; P=.007), and motor score (-2.8 [2.2]; P=.045); the Modified Rush Tic Rating Scale Score total score (-5.8 [2.9]; P=.01); and the phonic tic severity score (-2.2 [2.6]; P=.04). Continuous, off, and scheduled stimulation conditions were assessed blindly in an acute experiment at 6 months after implantation. The scores in all 3 conditions showed a trend for improvement. Trends for improvement also occurred with continuous and scheduled conditions performing better than the off condition. Tic suppression was commonly seen at ventral (deep) contacts, and programming settings resulting in tic suppression were commonly associated with a subjective feeling of calmness. CONCLUSIONS This study provides safety and proof of concept that a scheduled DBS approach could improve motor and vocal tics in Tourette syndrome. Refinements in neurostimulator battery life, outcome measure selection, and flexibility in programming settings can be used to enhance outcomes in a future larger study. Scheduled stimulation holds promise as a potential first step for shifting movement and neuropsychiatric disorders toward more responsive neuromodulation approaches. TRIAL REGISTRATION clinicaltrials.gov Identifier: NCT01329198.

Quantifying long-term microelectrode array functionality using chronic in vivo impedance testing

Long-term acquisition of high-quality neural recordings is a cornerstone of neuroprosthetic system design. Mitigating the experimental variability of chronically implanted arrays has been a formidable task because the sensor recording sites can be influenced by biotic and abiotic responses. Several studies have implicated changes in electrical interface impedance as a preliminary marker to infer electrode viability. Microelectrode impedance plays an important role in the monitoring of low amplitude and high-resolution extracellular neural signals. In this work, we seek to quantify long-term microelectrode array functionality and derive an impedance-based predictor for electrode functionality that correlates the recording site electrical properties with the functional neuronal recordings in vivo. High temporal resolution metrics of this type would allow one to assess, predict, and improve electrode performance in the future. In a large cohort of animals, we performed daily impedance measurements and neural signal recordings over long periods (up to 21 weeks) of time in rats using tungsten microwire arrays implanted into the somatosensory cortex. This study revealed that there was a time-varying trend in the modulation of impedance that was related to electrode performance. Single units were best detected from electrodes at time points when the electrode entered into the 40-150 KΩ impedance range. This impedance trend was modeled across the full cohort of animals to predict future electrode performance. The model was tested on data from all animals and was able to provide predictions of electrode performance chronically. Insight from this study can be combined with knowledge of electrode materials and histological analysis to provide a more comprehensive predictive model of electrode failure in the future.

Abiotic-biotic characterization of Pt/Ir microelectrode arrays in chronic implants

Pt/Ir electrodes have been extensively used in neurophysiology research in recent years as they provide a more inert recording surface as compared to tungsten or stainless steel. While floating microelectrode arrays (FMA) consisting of Pt/Ir electrodes are an option for neuroprosthetic applications, long-term in vivo functional performance characterization of these FMAs is lacking. In this study, we have performed comprehensive abiotic-biotic characterization of Pt/Ir arrays in 12 rats with implant periods ranging from 1 week up to 6 months. Each of the FMAs consisted of 16-channel, 1.5 mm long, and 75 μm diameter microwires with tapered tips that were implanted into the somatosensory cortex. Abiotic characterization included (1) pre-implant and post-explant scanning electron microscopy (SEM) to study recording site changes, insulation delamination and cracking, and (2) chronic in vivo electrode impedance spectroscopy. Biotic characterization included study of microglial responses using a panel of antibodies, such as Iba1, ED1, and anti-ferritin, the latter being indicative of blood-brain barrier (BBB) disruption. Significant structural variation was observed pre-implantation among the arrays in the form of irregular insulation, cracks in insulation/recording surface, and insulation delamination. We observed delamination and cracking of insulation in almost all electrodes post-implantation. These changes altered the electrochemical surface area of the electrodes and resulted in declining impedance over the long-term due to formation of electrical leakage pathways. In general, the decline in impedance corresponded with poor electrode functional performance, which was quantified via electrode yield. Our abiotic results suggest that manufacturing variability and insulation material as an important factor contributing to electrode failure. Biotic results show that electrode performance was not correlated with microglial activation (neuroinflammation) as we were able to observe poor performance in the absence of neuroinflammation, as well as good performance in the presence of neuroinflammation. One biotic change that correlated well with poor electrode performance was intraparenchymal bleeding, which was evident macroscopically in some rats and presented microscopically by intense ferritin immunoreactivity in microglia/macrophages. Thus, we currently consider intraparenchymal bleeding, suboptimal electrode fabrication, and insulation delamination as the major factors contributing toward electrode failure.

Input-output mapping performance of linear and nonlinear models for estimating hand trajectories from cortical neuronal firing patterns

Linear and nonlinear (TDNN) models have been shown to estimate hand position using populations of action potentials collected in the pre-motor and motor cortical areas of a primates brain. One of the applications of this discovery is to restore movement in patients suffering from paralysis. For real-time implementation of this technology, reliable and accurate signal processing models that produce small error variance in the estimated positions are required. In this paper, we compare the mapping performance of the FIR filter, gamma filter and recurrent neural network (RNN) in the peaks of reaching movements. Each approach has strengths and weaknesses that are compared experimentally. The RNN approach shows very accurate peak position estimations with small error variance.

Corrosion of Tungsten Microelectrodes used in Neural Recording Applications

In neuroprosthetic applications, long-term electrode viability is necessary for robust recording of the activity of neural populations used for generating communication and control signals. The corrosion of tungsten microwire electrodes used for intracortical recording applications was analyzed in a controlled bench-top study and compared to the corrosion of tungsten microwires used in an in vivo study. Two electrolytes were investigated for the bench-top electrochemical analysis: 0.9% phosphate buffered saline (PBS) and 0.9% PBS containing 30 mM of hydrogen peroxide. The oxidation and reduction reactions responsible for corrosion were found by measurement of the open circuit potential and analysis of Pourbaix diagrams. Dissolution of tungsten to form the tungstic ion was found to be the corrosion mechanism. The corrosion rate was estimated from the polarization resistance, which was extrapolated from the electrochemical impedance spectroscopy data. The results show that tungsten microwires in an electrolyte of PBS have a corrosion rate of 300-700 μm/yr. The corrosion rate for tungsten microwires in an electrolyte containing PBS and 30 mM H₂O₂ is accelerated to 10,000-20,000 μm/yr. The corrosion rate was found to be controlled by the concentration of the reacting species in the cathodic reaction (e.g. O₂ and H₂O₂). The in vivo corrosion rate, averaged over the duration of implantation, was estimated to be 100 μm/yr. The reduced in vivo corrosion rate as compared to the bench-top rate is attributed to decreased rate of oxygen diffusion caused by the presence of a biological film and a reduced concentration of available oxygen in the brain.