Flashcards in lecture 6- fMRI Deck (42):
why is PET not used?
- it involves administering a radioactive isotope to the patient (e.g. oxygen-15) thereby exposing the patient to a -non-insignificant amount ionizing radiation
Consequently, in the research setting, fMRI is now far more commonly used as it does not involve radiation
Basic Principles of fMRI
-It is a way of imaging the activity in the human brain
--Originally called NMRI (Nuclear Magnetic Resonance Imaging), but the term nuclear was thought to have too many negative connotations
fMRI versus MRI?
-fMRI images the activity of the human brain.
-MRI images the structure of the human brain (i.e. shows you where the skull is, where the white matter is etc).
-BOLD = Blood Oxygen Level-Dependent
-Neural activity uses oxygen
-Thus, when neurons fire, the brain increases the blood flow to them
-Because the brain sends so much blood to the active area, the oxygen content of the blood actually increases, the opposite of what you might expect
bold fmri continued
-Essentially one compares the BOLD fMRI signals coming from the brain in two situations 1) when the subject performs the task and 2) when the subject either does nothing or performs a control task.
-Subjecting (2) from (1) reveals those areas of the brain that were preferentially activated by the task
-In my fMRI experiment there were two conditions. In both, the observer saw exactly the same stimulus
-In Condition 1, the observer tracks targets
-In Condition 2, the observer passively views the same stimulus
-If brain area A is involved in tracking objects, then it will be more active in Condition 1 than in Condition 2
Consequently, its BOLD fMRI activity will be greater in Condition 1 than in Condition 2
-Thus, by subtracting the BOLD fMRI activity of Condition 2 from Condition 1 I could identify those brain areas involved in object tracking
basic principles 2
-The brain contains hydrogen atoms
-These nuclei of these hydrogen atoms (i.e. the protons) act as little bar magnets
-When the protons are placed in a (very strong!) magnetic field, they attempt to align with the magnetic field
-However, they don’t do this perfectly
-As the proton precesses, the direction of its magnetic axes changes
-The frequency of the pecession is the resonance frequency of the proton
Basic priciples 3
-A radio frequency (RF) pulse is supplied while the tissue is in the magnet
-If the resonance frequency of the proton matches the frequency of the pulse, the proton will absorb the energy and “flip” to a higher energy state
-This is why there is an “R” in fMRI
-proton will absorb energy only if it is also at its resonance frequency.
-When the radio frequency pulse is turned off, the protons will remit their stored energy, generating radio frequency pulses.
-The computer decodes the pulses to create an image of the brain
1st crucial fact
1st- not all the stored energy is emitted immediately.
It takes time for the protons to “flip” back to their aligned states
As they flip back they emit a signal
2nd crucial fact
-The precession frequency of proton depends on the strength of the magnetic field.
-Thus, if different protons are in different magnetic fields, their precession frequencies will be different.
-Thus, if the magnetic field is inhomogeneous, neighbouring protons get out of phase, so the RF signals they emit will also be out of phase.
summary of basic principles
-If the magnetic field in inhomogeneous, different protons will precess at slightly different frequencies.
-Thus, the signals from different protons will get out of phase with each other and so begin to cancel each other out.
-The more inhomogeneous the magnetic field, the faster the signals from different protons will become dephased, the more mutual cancelation will occur, the faster the signal will decrease.
-Thus by measuring the rate at which the strength of the RF signals emitted by the tissue decreases we can estimate how homogenous the magnetic field is.
-It turns out that the higher the blood oxygen level, the more homogenous the magnetic field, thus the slower the RF signal emitted from the brain decreases.
take home message basic principles
-we first excite the brain with an RF pulse.
-Then we measured the RF pulse emitted in turn by the brain.
-By measuring how long the brain’s RF pulse takes to decay we can infer the neural activity in that region of the brain.
-The longer the decay rate, the greater the neural activity
How Does fMRI Excite Just Part of The Brain?
-If the scanner were to excite the entire human brain at the same time, then we would not be able to figure out from which parts of the brain the result signals originated from.
-Instead, only one slice of the brain is excited at a time.
Consequently, we then know that the resultant signals must have originated from the slice that was excited.
-But how can you excite just one slice at a time?
Mechanisms of Scanning
-protons will absorb radio frequency pulses only when the frequency of the radio frequency pulse matches the protons precession (i.e. resonance) frequency.
-Thus, by causing the magnetic field to vary linearly, we can cause the resonance frequency to vary throughout the brain
-Thus, an RF pulse will excite only a slice of the brain – that slice of the brain where the resonance frequency of the protons matches the frequency of the RF pulse
-In this way, one slice of the brain can be selected at a time for imaging
-By imaging the brain one slice at a time, a 3D image of the brain can be created
-Typically it takes about 2-3 seconds to obtain a full scan of the brain in this manner
Measuring The BOLD Signal Within The Slice
look at lecture slides
summary for measuring slices of the brain
-We excited the brain with an RF pulse and then measured the resultant RF pulse emitted by the brain.
-Because we excited only one slice of the brain – we knew the z-coordinate of all the resultant RF signals
-We then briefly varied the magnetic field in the x-direction, so that the phase of the nuclei precessions varies in the x-direction. Thus by measuring the phase of the brain’s RF pulse we could determine the x-coordinate of the signal. This is known as phase encoding.
-During readout we varied the magnetic field strength in the y-direction. By listening to just one frequency, you can record signals of a particular y-coordinate.
summary brain slice 2
Slice-specific excitation: z-coordinate
Phase encoding: x-coordinate
Frequency-specific readout: y-coordinate.
There are a number of versions of this technique
What Does Bold fMRI Actually Measure?
-is a measure of the oxygen level in the blood
-This is influenced by the neural activity but not entirely determined by the neural activity
-Thus, one has to be careful when trying to infer neural activity from the BOLD signal
-Perhaps the biggest concern is that there is a hemodynamic lag between the increase in neural activity and the increase in the BOLD signal
the peak BOLD response is delayed by about 8 seconds and continues for about 15 seconds after the stimulus has disappeared
-Maximum temporal resolution typically determined by hemodynamic response function.
-Maximum temporal resolution is about 2 seconds.
-Thus, if two neural events occur within about two seconds of each other, the BOLD fMRI responses to the two events will become confused.
-The upper limit of the spatial resolution achievable by the BOLD technique is determined by how accurately the body can direct blood to the the volume of the brain that needs it.
-If a given neuron “requests” more blood, all neighbouring neurons within about 1 mm of it will received an increased blood flow.
-Thus, if two neural events occur within 1 mm of each other, BOLD fMRI will not be able to distinguish them.
-For practical purposes, one typically scans at a lower resolution with voxels of 3x3x3 mm3
-Because fMRI is non-radioactive it has no long term effects on the subject and the subject can be scanned repeatedly
-The only major safety concerns are due to proximity of a very strong magnet
-Typically the magnet is 1.5T – 3T (i.e. 30,000 to 60,000 times the strength of the earth’s magnetic field)
-Any ferromagnetic object brought too close will literally fly into the magnetic, destroying whatever is inside the magnet (e.g. the patient)
Kanwisher et al. (1997)
-Was the first study to demonstrate that there is a specialized part of the brain for processing faces.
-This demonstrates that not all objects are processed the same way.
conditions in Kanwisher 1997
Condition A: presented subjects with faces
Condition B: presented subjects with pictures of objects
Found areas of the brain that responded more strongly to faces than objects
what area responds to faces?
Found that an area of the fusiform gyrus responded more strongly to faces than to objects.
Now days, this area is referred to as the fusiform face area
did kanwisher redo the experiment?
yes found same result across subjects and using different control stimuli
summary of kanwisher 1997
Specifically, Kanwisher et al. showed:
-Their results are repeatable within a subject
-Their results are repeatable across subjects
-The FFA responds more strongly to faces than scrambled faces
-The FFA responds more strongly to faces than houses
-The FFA responds more strongly to faces than hands
Common Criticisms of fMRI Investigations:Multiple Comparisons Problem
-In fMRI one considers each voxel in turn and performs a t test to compare the activity in condition A versus the activity in condition B.
-For example, in my object tracking study there were two conditions. In both, the observer saw exactly the same stimulus
In Condition 1, the observer tracks targets
In Condition 2, the observer passively views the same stimulus
For each voxel in turn, I compared the fMRI activity generated by the two conditions via a t test
multiple comparisons 2
-The areas in red/yellow are those where t tests applied voxel by voxel showed that the activity in the tracking condition (condition 1) was significantly greater than the activity in the passive viewing condition (condition 2)
-In other words, for these areas, I used t tests to reject the null hypothesis that the activity in the two conditions was equal in favour of the hypothesis that the activity was greater in condition 1.
multiple comparison 3
The problem is that the more t tests that are performed, the greater the possibility of a false positive (i.e. the greater the probability of incorrectly rejecting a null hypothesis).
For example, if you were to perform 2 independent t tests at the 1% level, there is an approximately 2*1% = 2% chance that you are going to report at least one false positive
multiple comparisons simply defined
This is known as the multiple comparisons problem: the more tests you do, the greater chance there is of reporting a false positive
multiple comparisons 4
-In fMRI one divides up the brain into small 3D voxels (typically 3x3x3 mm3 cubes)
-As the total brain is approximately 1,500,000 mm3, this results in approximately 50,000 voxels
-Typically, in an fMRI study there are two conditions (call these conditions A and B) and one compares the BOLD -fMRI activity generated in these two conditions
Crucially, this is done by a t test on a voxel by voxel basis
-Thus, one is performing on the order of 50,000 t tests
-So, if you were to perform each t test at the 0.01 level of significance, you would expect 50,000 x 0.01 = 500 positive results even if the two conditions were not significantly different! (i.e. 500 false positives
-if you want there to be overall a 1% chance of reporting a false positive when you perform n t tests, then you perform each t test at the 0.01/n level
-For example, if you were to perform 2 t tests, you would perform each t test at the 0.005 level, so that overall there is only a 1% chance that at least one of the t tests will result in a false positive
-If you were to perform 50,000 t tests, you would perform each t test at the 0.01/50000 level, so that overall there is only a 1% chance that one of the t tests will result in a false positive
what is the issue with multiple comparisons?
-The problem is that some of the earlier fMRI studies did not correct for multiple comparisons or did not do so adequately.
-For example, Culham et al (1998) did not correct for multiple comparisons at all
-Thus, we do not know which of the activations she reported are false positives.
-In order to avoid the multiple comparisons problem, many studies specify a region of interest (ROI)
-By averaging all the voxels within an ROI, they do just one t test per subject.
-Thus, they do not have to perform the Bonferroni correction so they are more likely to find a statistically significant result.
Non-Independent Sampling 2
-Think of an ROI as just a big voxel.
-You select a bunch of individual voxels and then collapse them together to form just one big voxel (the ROI).
-You then do a t test on this big voxel comparing its activity in the two conditions.
-Because you are doing only one t test per subject you do not have to correct for multiple comparisons.
-You are less likely to miss a statistically significant result (i.e. you are less likely to make a type 2 error).
non independent sampling confound
non independent sampling is a valid method if the ROI is chosen in one scanning session but the data used for the t test is taken from a second scanning session.
-In other words, selection and testing must use different data sets (i.e. use different scanning sessions).
-Unfortunately, many studies use the same scanning session to select the ROI’s and to test the ROI’s. This is known as the non-independent sampling confound.
Biased Sample Selection
Grill-Spector et al. challenged conventional wisdom and suggested that FFA actually process more than just faces
biased sample selection 2
-She presented the subject with pictures of faces, animals, cars and sculpture
-She identified which voxels in FFA responded maximally to each category and created 4 regions of interest (ROI) within the FFA: one for faces, another of animals, another for cars and another for sculptures
-So, for example, the ROI for faces contained all those voxels that responded more to faces than to any of the other 3 categories.
-Within each ROI, she then compared the BOLD fMRI activity generated when she presented each of the four category types compared to when she presented random noise.
Thus her method of selecting the voxels was not independent of the method used to test the voxels.
Non-Independent Statistical Tests
-She claimed that this showed that different regions of FFA were highly sensitive to different objects (e.g. the car ROI was responded very strongly to cars but not as strongly to faces, animals or sculptures)
-Of course, this reasoning is invalid. Her sample was biased, so her statistical tests are invalid
The reason why the car ROI responded so strongly to cars was that this is how the voxels were selected!
-To demonstrate the statistical problems inherent with the Grill-Spector et al. study, Baker et al. redid the study, using the same flawed technique, and obtained the same results.
-They then did the same analysis, but now considered voxels that were not in the brain!
Again they got the same results!
-Baker et al. then choose the ROI’s using the data from one scanning session and tested the ROI’s using data from a second scanning session
how common is non-independent sampling problem?
In a recent survey of 55 fMRI studies in high impact journals, it was found that over half made this error
Over Interpretation of Null Results
-In fMRI, for most voxels there will be no statistically significant difference between the two conditions.
-It is said that for these voxels that you found a “null” result (a null result is just a result that was not statistically significant)
-What can you conclude about these voxels?
Answer: Not much.
-It could be that for these voxels there really was a difference between the two conditions but that the statistical tests were not sensitive enough to detect it.