Identifying Intra-Cortical Recording Instabilities

Maran Ma, Theodoros P. Zanos, Matthew R. Krause, Christopher C. Pack and Timothy E. Kennedy

Department of Neurology and Neurosurgery, Montreal Neurological Institute, McGill University, Montréal, Canada

1 OBJECTIVES

Cortical multielectrode-arrays (MEA) offer some of

the highest resolution technology for detecting

clinical user intent for controlling prosthesis, such as

robotics and functional electrical stimulation. The

long term performance of these neuron-machine-

interfaces (NMI) depends on both neural directional

tuning stability and signal recording stability (Wang

et al., 2014). Having more functional channels to

decode from a larger neural ensemble however, can

effectively compensate for directional tuning

fluctuations (Nuyujukian et al., 2014).

To improve MEAs to retain more stable channels,

it is critical to prioritize engineering requirements, as

the biological vs. non-biological causes of instability/

failure may require opposite design strategies. For

example, reducing gliosis may necessitate enhanced

biomimicry and softer substrates, while reducing

hardware failure may benefit from adsorption-

repelling and robust substrates.

This project aims to discern hardware vs. tissue

degradations that underlie recording instability, and

quantify MEA stability without relying on spike

sorting – as sorting is not essential to NMI, labour

intensive and/or has uncertainies (Einevoll et al.,

2012). To accomplish this we analysed chronic Utah

array recordings from inferotemporal and prefrontal

cortices of adult macaque monkeys, focusing on

individual channel trends.

2 METHODS

Existing recordings obtained from eight-months of

visual system studies on an adult macaque monkey

(monkey F) was used. MEAs were Iridium oxide

Utah arrays implanted in the left inferotemporal

(implant 1) and right prefrontal (implant 2) cortices.

Before each experiment an 8-minute baseline

recording was made with the same series of visual

stimuli (10 repetitions of 100 images per session) and

fixation point as previous sessions. These baseline

recordings are used in our stability analysis.

Data collected at 30Khz was digitally filtered at

250Hz-7.5KHz and thresholded at -4.5RMS for spike

detection. Channel spike rate distributions were

tabulated per session, fitted to gamma distribution,

and verified with chi-square goodness-of-fit test.

3 RESULTS

3.1 Stability Quantification

The most active channels were selected (based on

mean spike rate over all sessions), and analysed

individually. To examine channel data in detail, spike

counts within each 8-minute recording was tabulated

per second into histogram form (Figure 1). Since the

stimuli sequence included a wide variety of images,

each distribution represents a general range of

behaviour of the neurons at that electrode. If these

ranges change significantly between sessions (under

the same stimuli) the channel’s neural population has

likely changed, and is defined as unstable.

Figure 1: Spike rate histograms per session from channel

23, which showed notable discrepancy between sessions.

The spike rate histograms were found to fit

gamma distributions. Since each session contained 10

repetitions of identical stimuli, distribution

parameters extracted from single repetitions could

serve as a standard for nominal variability within a

stable period. This enables, for example, comparing

Ma, M., Zanos, T., Krause, M., Pack, C. and Kennedy, T.

Identifying Intra-Cortical Recording Instabilities.

In Extended Abstracts (NEUROTECHNIX 2016), pages 14-15

the parameters’ percent change between sessions vs.

within sessions as a criterion of channel stability. We

intend to further this analysis and quantify stability

over time for both implants.

3.2 Biological vs. Non-biological Causes

Intra-cortical recordings consist of high-amplitude

action potentials and a dense core of device and

biological noise. The latter is the larger component

(Lempka et al., 2011), and contains the overlapping

activity of a sea of distant neurons. Hence if the loss

of spike detection is due to hardware degradation, a

change should also be reflected in the below-

threshold portion of the recording. Inspection of

representative sessions (Figure 2) show that when

spikes are absent, the core portion is unaffected, and

signal loss is reversible. Thus hardware failure is not

the cause of this form/instance of instability.

Figure 2: Time domain signal from channel 23, sessions 16

to 18; red line is the spike detection threshold (-4.5RMS).

Histograms of waveform amplitudes distinguish the

bulk that is biological noise (Figure 3). Measuring the

low-amplitude side of this distribution quantifies

noise consistency, hence hardware stability across

sessions. A simple colour map visualization of all

sessions also implicates instances of hardware failure

(unusual distributions at the core signal component).

Figure 3: A) Un-thresholded signal amplitude histogram

showing noise consistency regardless of spike stability. B)

Top view of the stack of histograms for channel 23.

4 DISCUSSION

It is helpful for new NMI design and evaluation to

have simple methods to assess recording stability and

distinguish tissue or hardware induced signal loss.

Typically, stability is quantified from combined

activity of all channels. Such statistics are compelling

but hide the profile of single channels, which suggest

underlying mechanisms of failure. Also common is to

apply thresholding to eliminate noise and isolate units

prior to analysis, but this imposes a zone around the

electrode where only neural activity within is kept as

data, and more distant activity is discarded as noise

(or heavily filtered for LFP). This "biological noise"

holds robust information on hardware performance

unaffected by the state of local neurons.

We examined long term intra-cortical recordings

from sensory & processing areas of macaque cortex -

where neural activity can be regulated by stimuli - and

analysed individual channel spike rates and biological

noise: 1) Quantifying and fitting the distribution of

spike rates per session gave an illustrative statistic of

channel stability. This method does not rely on spike

sorting, which is non-trivial to perform and the

margins of error are harmful to stability assessment.

2) Tabulating all peaks in the recording showed that

the "core" of the amplitude distribution (comprised of

biological noise and system noise) can remain

constant when the spike count is low, indicating

cellular causes such as poor health of a neuron. This

analysis also identified sessions with abnormal signal

core shape, suggesting true hardware failure.

REFERENCES

Einevoll, G. T., Franke, F., Hagen, E., Pouzat, C. & Harris,

K. D. 2012. Towards reliable spike-train recordings

from thousands of neurons with multielectrodes. Curr

Opin Neurobiol, 22, 11-7.

Lempka, S. F., Johnson, M. D., Moffitt, M. A., Otto, K. J.,

Kipke, D. R. & Mcintyre, C. C. 2011. Theoretical

analysis of intracortical microelectrode recordings. J

Neural Eng, 8, 045006.

Nuyujukian, P., Kao, J. C., Fan, J. M., Stavisky, S. D., Ryu,

S. I. & Shenoy, K. V. 2014. Performance sustaining

intracortical neural prostheses. J Neural Eng, 11,

066003.

Wang, D., Zhang, Q., Li, Y., Wang, Y., Zhu, J., Zhang, S.

& Zheng, X. 2014. Long-term decoding stability of

local field potentials from silicon arrays in primate

motor cortex during a 2D center out task. J Neural Eng,

11, 036009.

Identifying Intra-Cortical Recording Instabilities