NEURAL CODE

Question 1

Intracellular recordings

Answer 1

– Action potentials from targeted cell
– Subthreshold membrane potential fluctuations

Question 2

Extracellular recordings

Answer 2

– Action potentials (spikes) from nearby cell(s)
– Sort spikes based on, e.g., shape, to individual cells
– Subthreshold fluctuations summed from nearby cells
– These are called “local field potentials” (LFP)

Question 3

Size and shape of electrode contact or tip

Answer 3

– A small exposed metal contact or tip of electrode has a high resistance (harder for currents to flow through)
– Smaller the exposed metal contact or tip of electrode, smaller the brain area we sample

Question 4

Electrode impedance

Answer 4

– Impedance is a measure of resistance plus electrode capacitance (avility to store change)
– Smaller the electrode contact or tip, higher the electrode impedance

– E.g., fine metal tip with only a few microns exposed metal would have a high impedance
(>1 megaohm) and be able to isolate spikes from individual neurons
– Fine metal tip electrode needs to be within 10’s-of-microns from neuron to record spike
– E.g., typical ECoG surface electrode with a larger exposed metal contact might have a
lower impedance (around 0.25 megaohms) and not be able to isolate individual neurons

Question 5

Local field potential (LFP) from extracellular depth electrode

Answer 5

– Reflects, e.g., up to 1,000-ish cells
– Derived mainly from within 250 microns of electrode tip

Question 6

LFP from electrocorticography (ECoG)

Answer 6

– Intracranial recordings from epilepsy patients
– Performed to localize seizure activity (but also research)
– Electrodes on exposed brain surface (subdural)
– Derived mainly from superficial layers of cerebral cortex

Question 7

EEG signals

Answer 7

– Reflect, e.g., 100s-of-thousands to millions of cells
– Summation of synchronized activity of neurons
with similar spatial orientation
– Predominantly derived from pyramidal cells in cortex
– Electrodes above scalp (arranged in cap for ease of use) i.e., non-invasive
– Skull smears EEG signal, degrading source localization
– Deep brain structures inaccessible to EEG
– Poor spatial resolution, but good temporal resolution

Question 8

Functional magnetic resonance imaging (FMRI)

Answer 8

– Excite hydrogen atoms with magnetic fields
– Measure emitted radio frequency signal
– Indirect measure of neural activity
– Blood oxygen level dependent (BOLD) changes with neural activity
– Reflects subthreshold membrane potentials
– I.e., better correlated with LFP than spikes
– Spatial resolution, e.g., 2x2x2mm3
; better than EEG
– Poor temporal resolution (e.g., sample every 2s)

Question 9

Voxels

Answer 9

– Brain subdivided into cubes called “voxels”
– Common voxel size 2x2x2mm3 or 3x3x3mm3
– ≈100,000 voxels common for FMRI study

Small effects, e.g., 2% change in signal

Question 10

Spike rate code

number of spikes in a given interval

Answer 10

– Much, much evidence for rate coding across the brain
– Generally speaking, increasing stimulus intensity, increases number of spikes (up to a point)

Question 11

Pooled response code

number of spikes from multiple cells in a given interval

Answer 11

– Combining activity from many cells reduces “noise” from variability of individual cells

Question 12

Labeled-line code

Answer 12

Vector formed from joint firing of multiple neurons

Which neurons fire as well as the number of spikes is important here

Question 13

Potentially more information in spike train than just number of spikes?

Spike train is a series of spikes

Answer 13

– Spikes do not always re-occur after a fixed time, i.e., there is variability in spike timing
– Is it simply “noise” in the system?
– Or could it be useful information?

Question 14

Spike timing codes

Temporal codes

Answer 14

– Spike pattern code
– Spike-phase code

Question 15

Spike timing codes

Answer 15

Spike pattern code, i.e., temporal pattern of spikes in a given interval

Each interval is divided into several smaller time bins

Question 16

Spike-phase code

Answer 16

Network oscillations provide a temporal reference frame or clock

e.g., spike timing relative to phase of oscillations

Question 17

Pattern classifier

Answer 17

Algorithm using multivariate neural activity to predict
what image category or class present at time of recording

Question 18

Training Step for Decoding

Answer 18

– Use subset of data to train classifier
– Classifier learns relationship between pattern of neural
activity and experimental condition (category or class)
– Linear and non-linear classifiers (dashed lines in figure)

Question 19

Test Step for Decoding

Answer 19

Classifier predicts category in which new data belongs

Question 20

FMRI data

Answer 20

– Generally need to average over many trials
– Decoding FMRI data from a single image
presentation significantly reduces accuracy

Question 21

What neural signal to measure?

Answer 21

– Spikes, but requires invasive technique
– LFP, but requires invasive technique
– FMRI is non-invasive, but not portable
– EEG is non-invasive and portable

Question 22

Neural prosthesis

Answer 22

Requirements:
– Stable long-term neural recordings from large numbers of neurons
– Efficient (real-time) computational data analysis
– Brain plasticity to incorporate feedback from effector (e.g., brace)