neuroimaging Flashcards
fMRI terminology :
session
run
block
experiment
session: single participant time into the fMRI in a day without leaving the machine
can have multiple runs
run : single fMRi recording ( 5-12 min )
can have multiple blocks
block : fMRI recording to a specific continuous condition
Advantages of fMRI
higher spatial resolution compared to other -_> allow for new types of analysis
non invasive –> high benefit low risk
easier to train people compared to other neuroimaging
anatomical voxel
smaller than the fucntion
spatial resolution, structural and functional voxels
spatial resolution : ability to detect differences, both structural and functional, across different location fo the map
for structural MRI
- spatial resolution means distinguishing afferent brainn areas ( the smallest structural change it can detect )
- voxel is smaller (0.05 to 1.5mm3)
pros: higher precision for restricted hypothesis
cons : low signal to noise ration, require more time for acquisition
functional MRI
- spatial resolution means the precision it can pinpoint activity on a map
- bigger voxel ( 2-3mm3)
pros: lower time , bigger noise to ration
cons: partial volume effect
partial volume effect
in fMRI bigger voxel can cause the machine to record acticity of voxels form different structural tissues, functional regions and slices os belonging to the same voxels
solution: use smaller voxels nd slices, use slower frequency
fictional resolution
the smallest detectable difference in two different bold activation
signal to noise ratio
the ration between activity in target area divided by external noise ( ex: physical motion)
contrast to noise ratio
the difference between two activated area divided by the (external ) noise
different within of firm have different contrats to noise ratios
dynamic Contrast to noise ratio OR functional signal to noise ratio
the difference within the same voxel in activation across different condition
task related variability / non task related variability
higher in sensory area rather than associative areas
SNR range
– Total range: 0.1 to 4.0 – Typical: 0.2 – 0.5
what to take into account for betewtr functional signal to noise ratio
since it is a measure depend experimental manipulation , make sure the manipulation is effective with the right stimulus –> ex: hone checking for the activation for FFA do not use the rest as control condition but use another visual stimulus
consider that different areas of the brain have different natural amount fo noise over time
thermal noise
variation detection of single due to fluctuation of electrons either in the field or in the machine
increases with temperature and field strength
truly random ( so can be decreased with averaging
mass motion ( mostly head )
if motion is larger than one voxel it is critical
motion artefact can be decomposed into 3 coordinates ( x,z,y with respect to anterior commissure ) and 3 movement components (and roll, jaw , pitch ) that can be used ars regressor of no interest to correct for the artifact
instructing the participant is fundamental ( can also have sham session to train) , also can be aided with some physical measure like padding
can also correct during preprocessing using realignment
inter subject variability source of noise
people differ in RT and also BOLD signal ( especially in some more evolutionarily new areas that present more connectivity difference across subject s)
validity problems as source of noise
RT is just a proxy
also different people might implement different strategies
general solution of increase signal to noise ratio
use PCA And ICA
use filtering ( careful it can cut out important frequency )
increase field strength ‘??????
averaging by the number of trials
problems with avaraging
- assumes noise is random - not correlated with other proccesses , but if not wea re ignoring potential relevant info
- we make activation cluster appear bigger
- since variability is actually divide by the square root of n of trial ( not just by pure n of trial ) at one point the difference betwween sir root of different numbers will not be detectable anymore , so advantage of averaging is lost
temporal resolution
ability to detect changes over time
problems with fMRI time resolutions
MEnon 1998 : some area might have fixed delayed onset times independently of the task type and duration
variability between subject –> averaging
variability between task in the same subject : make sure the difference in RT for different task is not due to confounding like duration of the task , difficulty , etc..
bold single is sluggish , it does not perfectly reflect neuronal activation times
dependency on sampling rate
how can sampling rate (repetition time ) affect time resolution
aliasing problems = different repetition time can give different bold signal
too short rt cudl take long time or have a small ccoverage
woudl noy giev the time to the flip angle to reach the peak –> smaller signal
too large RT can make the initial dip go undetected
possible solution = jittering : starting the RT at different latencies in the Bold signal timeline
what is the linearity system framework and what does really happen to bold signal of repeated stimuli ?
linearity systems assume
scaling : amplitude of single is proportional to the input magnitude
superimposition : the result of two successive stimuli is the summation of the two single output
BUT
we have refractory effects : the successive stimulus is influenced by the first one, t hey are not independent -> amplitude is lower and peak is later the closer they are
aka successive stimuli are under additive
how do refractory effects become useful in adaptation studies ?
ina adaptation studies , if adaptation is present the two stimuli are perceived as a single one , so there is only one response , and therefore no refractory effect
if adaptation ins not present ,we have two successive percieve stimuli , which will show refractory effects
ideal TR
for event related : 1.5-2 ms
for block design : 3-4 ms
preprocessing steps
visual inspection of scans and movies to detect mass motion
check
remove dummy scans
adjust distortion: use the map fo the magnetic field to adjust distorted images
realignment : superimpose images of same modalities from same participant , by using the 6 parameters values to apply rigid body transformation ( if leftover unpatching in activated areas , sue the parameter as noise parameters , so as regressor of no interest ))
sparse scanning : alternate scanning session with non scanning session –> damp the motion artefact by avoiding recording gin unrelevant period for activation
slice timing correction: interpolate interleaved recording as if happening at the same time ,
coregistration: different modalities scans form same participant are superimposed to increase spatial resolution
normalisation: estimate the normalisation parameters form the tame plate and then apply them to the normalised scans
smoothing : a sort of wighted averaging of each voxel signal with their neighbouring ones, depending on how much each voxel influence each other
pros: better superimposition ( even more than normalisation ), less number of comparisons since we are now comparing cluster of voxels not single ones, increase signal to noise ratio
cons: bad spatial resolution, bad for ROi analysis
General linear model components and general goal
Y= outcome aka activation of a voxel
x= predictors we take into account
b= how much of each parameters could be causing y
e= error
goal :
estimate for each voxel a pattern of predictors and estimate the value for their parameters such as the same pattern ( GLM ) can predict with different parameters, the activation of different voxels using different set of parameters for each voxel
How we do that ? just reminder Granziol : contrast matrix where we compare different time points ( rows) with different conditions ( columns ) to estimate the beta , then run test to first asses the model significancy and then the single parameters significancy
problems with GLM
- assume errors are normally distributed in each voxel and has the same variance in all voxels ( which sit not )
-does not take into account the fact that activation is not all or none –> usually we convolute an actual HRF activation with the one obtained by the model
- does not take into account physiological artefacts
- assumes orthogonality in parameters–> in real life they are usually not
voxels are not independent from each other–> we don’t actually have many single isolated activations ALSO blood flow ( the proxy we use ) is not really localised and spreads
problem with multiple comparison
the more test we run the higher the probability of false positive solely because we have more test
so we have to correct for the number of comparison AKA diminish the single parameter p value so that when taking multiple test the whole model p value is still reliable
main corrections for multiple comparisons
Bonferrroni :
divide the the individual parameter p value by the number of test ( problem if you do voxel by voxel comparison so with 50 to 100 thousands fo comparison
Family Wise Error Correction with gaussian field theory :
estimate a priori the probability of fall expositive for each voxel assuming they have a normal distribution of activation ( so kinda considers clusters )
False discovery rate : estimate , from the significant results, the probability of them being false positives
ROI
we estimate significant areas in that are signifiant and run a different set of test on that area specifically , reducing the number of voxel to compare
or
just take predefined areas and run test within them reducing automatically the number of voxels to compare
or we just compare rois
first order analysis
and second orders analysis
first ( fixed effects ) : assume inter subject differences are noise and that effect are the same for all subjects (aka Response magnitude is fixed)
secodmn ( random effects). take into account inter individual difference ( magnitude is random ) by using the map of individual activation to estimate a single map of activation
block design
pros:
no task switching effect
easier analysis
easier fro participants and tester
cons:
can make the subject predict the stimuli
some task cannot be bloked
cannot differentiate individual responses to specific stimuli
event related design
pros
allow for post hoc analysis ( go back and pin point the activity correspondent to a specific stimulus )
make up for blocked design cons
cons :
some processes happen online 8 ex gestalt ) and can’t be predefined in a n event related design
some processed make it difficult to switch an cannot be made into a strip t pattern
longer experiment
history of neurostimulation
galvani : found that muscles moved with electrical impulse, hypothesised myelin and ion channels
aldini : used electrical stimulation,ulation to treat melancholia
ferrier : stimulation of cortex revealed cortical maps ( effect in one cortex area showed result in a specific body part 9
MEDUNA : discovered elctroconvulsive therapie to treat some disorders like depression
penfield: homunculus and induced memories via electrical stimulation
Magnusson & Stevens . found that magnetic field applied by coil had no reaction when continuously used, but di when stopped or restarted
Bickford & Freeming : elicited muscles contraction with magnetic stimulation for peripheral nerve
advantages of ems
causal elation possible
located manipualtion
possibility to change control site and task
minimise the effect of plasticity any having shorter and non longitudinal l studies ???
can manipulate networks????
neural chronometry
disadvantage : those damn cortical areas
how does tms work
electric field in coil, elicit magnetic field that on skull elicit electric current on cortex
what to take into account
orientation of neuron can determine whether effect is hyper or de polarisation???
orientation of the coil
direction of the current
monophasic vs bifasic
monophonic ( current in one direction??)
is less powerful ( requires more intensity to reach same result as bifasic
easier to see effect of summation since affecting more specific neural population=????
bifasic ( current is a sinus )
can recover more fastly so more suited for high frequency pulse s
how to find the right areas to stimulate
stimulate an area and derive which one it is forms he effect fprovoked
uset he eeg 10/20 system
navigate form a know area if you know their relative position
use a neuonavigation map
how does fMRI work
protons that have moth a magnetic momentum and an angular momentum are said to have nuclear magnetic resonance (NMR) property
laos we need molecules that have an odd number of protons or they cancel each other out
hydrogen is one of those
we align the portions magnetic momentum by applying a constant magnetic field
then we tip the portions out of alignment using radiofrequencies at a specific frequency at which protons spin ( larmor frequency , calculated with angular and magnetic momentum values , protons of different molecules have different larmor frequencies)–> this make he potion go to a higher energy state
when they go back to alignement and to a lower energy state , they release the excess energy in the form of a oscillation , free induction decay , recorded by the rf coils
how hydrogen and deoxygenated and oxyhemoglobin relation on fMRI recording
even if we record hydrogen portions , the deoxygenated haemoglobin is a paramagnetic properties, modified the magnetic field, the oxyhemoglobin does not
so fMRi signal increase is dictated by oxygen -deoxy , with more oxygen increasing the signal
we know that blood flow is faster than oxygen consumption
so what happen when neural activity starts is that we have a higher need and thus supply of oxy, making the single increase ( we woudl expect a decrease in signal due to increase of oxeye consumption and thus increase of deoxygenated BUT NOPE)
Hemodynamic Response function has a canonical shape :
hypo oxygen stage : initial dip due to oxygen consumption before blood flow supply
hyper oxy stage : increase in signal due to arrival of oxy
peak
overshoot : oxygen consumption rate is catching up with oxygen flow , thus the ratio with deoxygenated is coming to a balance a making the signal decrease ( undershoot )
contrast agent ss pros and cons
they have higher paramagnetic properties than deoxygenated , so more sensitive
also the signal would now be based on blood volume not oxygen-deoxy ration
cons :
worse time course signal due to time requirements for the contrast agent to diffuse
participant can have side effects
injection of substances birth be discouraging
history of fMRI
Mosso : understood more activity in brain was linked to more supply of blood to brain , so patient ofn equilibrium woudl tip toward the head if activation–> not totally correct: increase in blood flow( what actually happens ) is not necessarily correlated to increse in volume so weight
Pauli : angular momentum
rabiu : magntic momentum
bloch and purcell : Determined relaxation times.–> They showed that energy applied at a resonant frequency was absorbed by matter, and the re-emission could be measured in detector coils–> NMR imaging
Lauterbur: NMR became visual –> Introduced magnetic field gradients (“spatial gradients”), which changed the spin frequency of atomic nuclei over space and thus allowed recovery of spatial information
DAMADIAN’: first NMR image
Mansfield: echo planar imaging –> getting hwole 2
ogawa : started functional MRI
functional contrast
measures differences in blood oxygenation (BOLD) between different confdition ( SNR is the actual value i think )
anatomical contrast
ability to distinguish between different tissue types (depends not only on voxel size!)
pros and cons of small voxel
better for restricted area hypothesis
higher spatial resolutions
lower signal to noise ratio
longer time
what re the CNR of T1 and proton density
t1:
high between grey and white
low between CSf and air
proton –> opposite
percentages of change in signal and in noise
singal : 0.5-2%
noise: 0.5-5%
what increases fucntional resolution but reduces spatial
smoothing
normalization
using ROI
scanner drift
could be due to gradual changes in inhomogeneities in the static magnetic field, or to slow head movement
use a high pass filter to correct it
Ghost artifacts
slight rhythmic movement like hear pulsation and berating
When the field of view (is not large enough to cover the entire area being imagedàsignals from outside FOV appearing on opposite side of image, creating “ghost” images
type of hypothesis regarding temporal resolution
type oen : detection–> is the signla there or not ?
type two : estimation–> estimate properties of the signal
shimming coils
reduce the static magnetic field inhomogeneities
radiofrequency coils
can be transmittent or receivers
surface coils are not transmitters due to high inhomogeneities but good receivers since they pic up signal only of area of interest adn spare extra noise form irrelevant areas
vlolume coils are both transmitters ( good for low inhomogeneities) and receivers ( ok but problem if we are interest in small area since it takes up al lot of noise )=
T1
longitudinal relaxation : the return of the magnetic vector along the static magnetic field ( along the z axis ) after it has been tipped away form it
t2
transverse relaxation:
after resonance , where proton spin in phase, t2 detect their return to out of phase spinning
t2*
is the transverse e relaxation that also takes into account the magnetic field disomogeneities (not only spin to spin interaction )
—> since the deoxygenated causes magnetic field disohomogeneities, the change in deoxygenated is perceived by t2–> t2 is the basis for bold detection
Total NMR signal
PREMSISE: The number of protons directly affects the strength of the NMR signal:
A higher proton density leads to a stronger signal.
Tissues with abundant hydrogen (e.g., water and fat) generate stronger signals compared to those with lower hydrogen content (e.g., bone or air
= depends on total number of protons reduced by t1, t2 and t2*
gradient coils
coils that creates predictable changes ( according to a pattern ) to magnetic field, so we get spatial resolution
relaxation times and color
t1: short relaxation time show as bright (fat/white =lighter) and longer as darker (less fat –> grey = darker , Csf = very dark )
t2: shot relaxation time show as dark ( fat / white matter ) and longer time as lighter (less fat and liquids )
oposite
ideal for maximum visual contrats ( since they have very different relaxation times
t1= grey -white
t2= tissue -csf
echo planar imaging
Kiev only one pulse and then rapidly shift t the gradient
what type of neural process dioes bold reflect more
NOT Sinaptic Density Function ( description of how the endogenous processe of the nuron affect activity– output descriptor ) or Multi Unit Activity ( combination of output of many neuron )
YES
local field potential , the sum of extracepllualr potential that get to the nuron–input descriptor
extra , bold signal does not linearly increase with stimulus intensity
How is bold connected to T2*
deixy emoglobina increases inhomogeneities–> faster dephasing of T2*–> lower BOLD signal
opposite
how long post stimulus does Bold reach peak ?
4-6 sec
what is the delay afte stimulus of bold signal ?
2 sec
how long is th total bold
16 sec fo even short stimuli
chat is the toal variance account by the generic HRF ?
70%—> this is why inter subject specific model are important, chi can account for 93% ( otherwise there would be no difference )
overshoot usually found in
bloked design
overshoot more common for
long stimuli (>10s)
initial dip and peak
initial dip is considered to be more spatially accrued as the onset of neural activity
peak still used as a proxy for onset of processing
Increasing the duration of neural firing (e.g., with longer stimulation)
ncreases HRF width
Increasing The Rate Of neuronal firing (e.g., with more intense stimulation)
ncreases HRF amplitude
Ogawa Et Al.,1992
they checked the signal at different echo time
since echo time only affected t2* and not t1 , if there were changes they would be due to t2*
the signal they saw at 80s and not 40 s was due to t2*
Blamireetal.,1992
easuring changes in visual cortex activity following stimuli of different durations
they saw that even stimuli as short a s2 sec woudl produce a significant change in the signal
what does realignement consist of
estimate the 6 parameters that woudl make the scan match onto the reference via rigid body transformation
pros and cons of coregistrationm
pro: we can take advantage of the structural images good spatial resolution basically
cons:
you cannot improve actually the spatial resolution of bad low res images
also we cannot make direct inference about the intensity sdiffercnes since they are taken with different components ( t1, t2, t2*)
two ways to normalise
- estimate parametrter to normalise from a t1 image and form a t1 template , then map it onto a functional image
- make a sort of average template from functional images , and then map each scan into that
tf is smoothing
smoothing takes the voxel of the image and substitute with the average of the voxel itself and the neighbouring voxels , those last each weighted base o nt he value of th eternal that we are using
statistical parametric mapping
eaning that a statistic (e.g., T-value) is calculated for every voxel – using the “General Linear Model”
assumes continuou data and normal distributed noise
the x matrix in the linear model
coloumns are condition ? or predictor?
and each row is a time point
so each whole column is a predicted BOLD signal course
good things about glm
can test complex hypothesis ,
can include variability in the data ( like head motion )
small volume correction
definbe an ROI and test just those voxels
can be defined a priori ( structural) or with statistical data ( functional roi)
WARNIGN : with functional roi do not use the same data used to determine statistically significant activation ROI to test for significance of the voxel inside the roi( of course they will be significant )
some components of SP M
mask.img= defines the area to look for activation
ResMS.img= info about variability of error
beta images= info about the betas estimates
beta images and resMS.img are combined to obtain the final image , the spmt.img
how to maximise t2
and t1
you need long TR , so to give time to t1 to fully longitudinal magnetization (M₀) recovery before the next excitation pulse.
also you need long TE so to give time to T2 relaxation to be completed
for t1 is the opposite , wee need short TR and short TE
the birth of fMRI with t2*
ogawa used t2* on rodents wo breathed 100% , 21% and =% oxygen , and 0% shoed up with dark spots where blood vessels were ( dark spots =deoxy)
describe neurovascular couplign
initially , due to neuronal activation , we have a compensatory response , where blood flow increases faster than the volume , flooding the area with oxy , then the volume adapts increasing , finally the flow returns to normal levels and the volume once again tales longer to adapt
fSNR =
task-related variability/non-task-related variability
what is a cost fucntion
is the function that tells us how much an image is different from the references (useful for realisgnement)
wha is mutual correlation
simile ro cost function but used when having different type of image to compare ( in coregistration) more similar to correlation
multicollinearity
ndependent variables in a regression model are highly correlated,
what is autocorrelated model
is a model that calculates the correlation of error and excludes it
categorical design
tests whether there is activation or not for a specific process
can use subtraction:
based on pure insertion assumption
we haev two task that differs for a component –> this mean if activation is present one y for the difference, thus the activation depends on that non overlapping process between the two tasks
ex: familias vs unfamiliar faces recognition
can use conjunction:
to minimise the problem o fvalididty of pure insertion
we isolate the process by making comparison between comparison
ex: color naming vs object naming AND object passive vision vs color passive vison–> you compare the are common between object naming and object passive vision not present in the other two category s
what are we testing for in the global null hypothesis
significant eset of effects
what are we testing for in the conjunction null hypothesis
set of consistently significant effects
parametric design
basically a simple lm model with just one predictor
we cehc k the change in signal depending on the change in one of our predictor, like intensity , number , duration etc.. rather than its mere presence
factorial design
looks for main effect and interaction
can be categorical ( lm with two categorical predictors) or parametric ( lm with one categorial and one continuous predictor I)
virtual lesion approach VS neuromodlatiuon approaches
virtual lesion aims at interfering with activity in one area so that the areas cannot perform its normal function
neuromodulation aimas at enhancing or inhibiting activity in one area so
type of pulses
single pulse: one pulse delivered per time , short lived effect
dual pulse: two pulses delivered one right after the other from the same could
multichannel /multicoil tms: using more than 2 coils ( 2 coils is the norm i think )
paired pulse : two pulses each delivered at a different location
example of neural chronometry with single pulse
disrupting V1 activity at different time points from stimulus presentation will tell us when V1 process onset is, cause we woudl have impaired fucntion at that time point with tMS use
example of neural chronometry with paired pulse
one stimulation at primary visual area ( static phosphene ) and one at v5( moving phosphene) —> it was found that when we stimulated v5 first and v1 after 20 ms phosphene was not seen moving , but static
AKA at that time V5 gives feedback that allow seeing movement
repetitive TMS
longer effect at cortical levels ( 100 ms ) –> does not need temporal precision–> not good for neural chronometry
can achieve longer behaviroal effects , usually lasting half of the total stimulation time
inhibitory and excitatory TMS frequencies
<1 are usually inhibitory
betwen 5 and 20 are usually excitatory
theta bursts are an exception ( inhibitory even if they are at high frequency range )
theta bursts
is at 5 hz , but each burst is made of 3 pulses at 50 hz
efefct last longer than the total stimulation time , up to 90 min
types of theta burst stimulation
intermittent : many bursts until they fill up 2 sec, then pause of 8 sec
excitatory effect on motor cortex
continuous : burst for up to 40 sec , inhibitory effect on motor cortex
types of threshold for pms intensity
individual : intensity to elicit response 50 5 of the time
motor: lowes intensity that elicits movements
phosphene: lowest intensity to elicit moving phosphene
what is the relation between motor and phosphene threshold
they are not correlated , so motor threshold cannot be used for visual areas stimulation and viceversa phospehen threshold cannot be used as a proxy for motor areas
yet phosphene threshold proved reliable for visual areas
what are the standard intensities ?
- single pulse: 60-80% maximum stimulator output (MSO) / 100-120% motor threshold (MT)
- repetitive: 50-70% MSO / 80-110% MT
- theta-burst: 35-50% MSO / 70-90% MT
what to know when using online tms
- When the processing starts
- How long it lasts
- Whether it is distributed/parallel over multiple areas
- Whether it is lateralized or bilateral
- Whether it is a feedforward processing or involves backward processing.
what can influence tms effects
the level of noise : certain amounts of tbs interference could actually be beneficial to some processes , and produce counterintuitive effects expected form lesion approach
neural state : studies report TMS to actually excite less active neurone , again counterintuitive effect to lesion approach
artifacts tms
clicks–> use earplugs or usi click also in control condition so ahabituation
vibration–> either vibration absorbing padding or give vibration in control as well
accidental cranial nerve stimulation ( can cause twitch and pain ) —> change the coil orientation ( but it woudl influence the effect ) or give tbs offline ( again , it changes the effect
what control could be used with tms
control site
control task
control stimulation ( sham )
problems with sham
cold be a placebo coil ,a 90° flipped one , or inverted, or with a panel between it an d the cortex
auditory effect is similar but the feeling of the sham does not resemble the one of real stimulation
problems with choosing control site
should not be structurally or functionally connected to teh target area
usually th homologous area is chosen
the heating problem
we either change the coil, cool the coil, use self cooling coils , or include this problem in the experimental paradigm
possibile side effects of tms
seizure( usually in patient s)
hearing loss
burns
effect on mood and cognition
neck and head ache
what and who should avoid TMS
pregnant people
brain injury
stroke
seizure
substance abuse
metallic medical aid
the levels of risk with TMS
1 : benefit is expected directly just from teh stimulation– risk is accepted
2. potential no verified benefit–> low risk
3 no benefit , just understanding –> needs to eb very very safe
types of tES
transcranial direct current stimulation: cstant current of 1-2 mA
t alternating current s: current flow is not continue can have different frequencies
t random noise s: current alternates randomly
t pulsed current s :
what does TES do
does not makes action potential fire but changes the level of excitability of neuron
anodal vs cathodal stimulation
anodal stimulation causes increased neural excitability ( depolarization ) → increases BOLD signal ( laso increased blood flow with PET )
cathodal stimulation (usually) causes decreased excitability. ( hyoerpolarization) –> decreases BOLD singal ( also decreased blood flow with pet
ACTAULLY Low intensity (1 mA): differential effects (anode vs. cathode)
Higher intensity (≥ 2 mA): increased excitability from both
anodal and cathodal polarity (Batsikadzeet al., 2013).
tes advanatges to TMS
cheaper
easy to carry and to use
acn use sham efficiently
no noise
