Lecture 2 - Neural measurments and analysis techniques Flashcards
Looking inside the brain
When using techniques to ‘look inside’ the brain you want to optimize a few aspects, namely:
- Coverage: The area you can measure with the specified technique.
if you can only measure a small spot -> low coverage, and vice versa.
- Temporal resolution: How precise your measurment of time is.
Different techniques have different order of magnitude for their measurments. - Spatial resolution: How precise your measurments of space is.
Different order of magnitude for their measurments.
Mobilty: Allowing the patient to still move while the experiment is ongoing.
For example, MRI has low mobilty.
Non-invasive: Very precise measurments often require invasive techniques in which components are placed
inside the patient’s head.
As this is not ethical, you should search for non-invasive techniques.
*The wish list for the perfect technique will be high coverage, high temporal, high spatial, high mobilty, and not invasive.
This is however not possible, since you always trade-off something for other advanteges or benfits. (e.g. high temporal resolution often leads to invasive techniques).
Marr’s level of explanation
helps ask the right question to see if a specific theory is implementable or not.
- Computational: why do things work as they do?
- Algorithmic: what representations can implement such computations?
- Implementational: how can such a system be built in hardware?
MEASURING NEURAL ACTIVITY (NON-INVASIVE)
As the duration of neuron firing is very short and neuronal currents are
very small, we need many neurons firing simultaneously, to measure
outside of the skull.
▪ Due to the hyperpolarisation phase of action potentials, they are very
hard to measure outside of the skull, as several action potentials appear to average each other out. Also, the duration of these potentials is very
short. If a primary current in the brain occurs, a secondary current will occur to bring the electrical potential level back to baseline. These are two types of currents we can measure. If you were to measure primary currents without technique, you would encounter the aforementioned problem as well, since
the primary and secondary currents average each other out. To measure neuron firing, we are mainly picking up the activity of post-synaptic potentials from pyramidal neurons.
ELECTROENCEPHALOGRAPHY (EEG)
▪ EEG is biased towards secondary currents, as it is
measured from the skin and the secondary currents
are the only currents to reach this layer.
- Every layer (brain, CSF, skull, scalp) has its
own conductivity which means that the signal
might sometimes stay within a specific layer
or be bounced back. This can affect signals,
as they might not fully reach the skin (scalp)
layer.
▪ EEG measures differences in electric potential at the scalp.
▪ EEG has a high temporal resolution (ms)
▪ EEG can see (deeper) radial and tangential sources.
▪ EEG has poor spatial resolution
▪ EEG requires substantial participant preparation time
MAGNETOENCEPHALOGRAPHY (MEG)
▪ MEG tries to measure minute changes in the magnetic field around your head. You need about 5k to 10k cells to fire together, to be able to measure the effect.
- As the magnetic field produced is very small, it
will need to be amplified and sensors that can
pick this up are required. This can be done by
using super-cooled SQUIDS in liquid helium,
that can be found in the MEG machine.
- A SQUID (superconducting quantum
interference device) is a very sensitive magnetometer used to measure extremely subtle magnetic fields.
- The magnetic field passes the scalp level and can therefore be picked up by the SQUID.
▪ MEG are biased towards primary currents of tangential dipoles, yet cannot pick up primary
currents of radial dipoles.
▪ MEG has high temporal resolution (ms)
▪ Deeper sources are less visible
▪ MEG signals are not affected by differences in conductivity (brain, CSF, skull, scalp)
▪ MEG has higher spatial resolution than EEG (but still low)
ELECTROCORTICOGRAPHY (ECOG)
▪ ECoG is similar to EEG, but the electrodes are
invasively placed on the surface of the brain and the
electrical activity is then recorded. This provides an
improvement in spatial resolution.
▪ High temporal resolution (ms) and a spatial resolution of about 1 cm.
▪ Grids can be slid under the dura matter to reach areas currently not exposed.
▪ It is only used for medical reasons (epilepsy).
MEASURING NEURONAL ACTIVITY VIA BLOOD FLOW
▪ When a brain region activates, there is an increase of blood, oxygen and glucose to that
particular region due to neurovascular coupling (the relationship between neuronal activity
and the required amount of influx).
- This changes the blood’s constituents, e.g. the amount of oxygenated hemoglobin.
This can be tracked as HbO2 has different magnetic properties and looks different.
FUNCTIONAL MAGNETIC RESONANCE IMAGING (FMRI)
▪ fMRI measures brain activity by detecting
such changes in blood flow.
▪ fMRI has high spatial resolution (mm)
▪ Due to the ‘sluggishness’ of blood, fMRI has
poor temporal resolution.
▪ Capillaries penetrating the grey matter have
very high spatial resolution.
▪ No problems with depth, but problems near
cavities.
FUNCTIONAL NEAR INFRARED SPECTROSCOPY (FNRIS)
▪ During fNRIS, near infrared light passes through the skull to measure the colour of the blood
in the brain. This is done by inserting a light injector.
▪ By using light at two or more wavelengths
the relative distribution of Hb and HbO2
can be measured by a light detector
▪ Using this method decreases the spatial
resolution immensely, and the temporal resolution is still rather slow.
▪ An advantage is that it combines well with
EEG to get multiple measurements.
STUDYING BRAIN ACTIVITY
▪ Many analysis of brain data reveal that when performing a specific task, the effects of such a
task are relatively small compared to the overall activity.
- This could lead to a larger inferential distance (gap between the person trying to explain and the person trying to understand), as there can be a lot of activity during a task, but how much of it is actually explained by the task?
▪ Many sources of variation are outside of our classical experimental paradigms.
- If you were to experiment on a mouse, it would not only move when instructed but perform all sorts of
(uninstructed) movements. Therefore much of the
variance (that is not caused by the task) in brain data can be explained by such uninstructed movements. Are we measuring the right dimensions of data?
▪ To what extent can data retrieved in a laboratory experiment be considered a natural
phenomenon? What is the ecological validity?
▪ Brain science is complex, and analysis pipelines are intricate including many (subjective) choices, which can lead to differences between researchers even when analysing the same dataset. Researchers tend to be optimists.
ANALYSIS TECHNIQUES
▪ While performing univariate analysis on fMRI brain data (looking at one voxel at a time), the
first view on visual representations would be that modules are selective to few distinct categories.
- Each voxel is a cube of pre-set dimensions (e.g. 3 by 3 by 3 mm), where you get the BOLD (blood-oxygen-level-dependent) activation from.
▪ While the spatial-average activation of a specific brain region might be of similar levels as
another, the activity patterns (vectors or points in high-dimensional activation space) might vary. For similar conditions such patterns will be close by in activation space, and for very different conditions far away.
- Looking at patterns and multiple voxels is called
multivoxel/multivariate pattern analysis. This focuses on
information rather than average signal. Signals from all
dimensions are used independently but we do not perform multiple comparisons.
- You can determine how different patterns are from each other
(and if they’re separable) with several methods, in which you try to separate the two patterns by finding a separating hypoplane.
▪ Representational Similarity Analysis (RSA):
taking two stimuli, measuring the patterns for
the stimuli and then how distant they are from
each other. This can be plotted in a matrix,
which will then tell you each power-wise
distance from one stimulus response to another
stimulus response (to see if things group or not).
