Detecting changes in brain activity using fMRI

1
Detecting changes in brain activity using fMRI
2
Questions for psychiatric fMRI

• What is the functional anatomy of disorders,
  symptoms and subsyndromes?
• Can we image functional effects of single genes
  or heritability in general?
• Can we image normal neurodevelopmental processes,
  normal adult brain variability, and understand
  pathogenesis in that context?
• Can we image functional effects of psychotropic
  drug action? do these predict clinical or
  cognitive benefits in individual patients? can we
  use fMRI to accelerate psychopharmaceutical drug
  development?

3
What does fMRI measure?

• fMRI is an indirect method of measuring brain
  activity.
• It does not look at the electrical activity of
  neurons directly.
• Instead it measures the way in which the
  circulatory system responds to the increased
  energy demands of activated brain cells.

4
The BOLD effect

• Most fMRI experiments use the so-called BOLD
  effect ( Blood Oxygen Level Dependent Contrast ).
• This is a small ( less than 5) change in the
  brightness of fMRI images resulting from a small
  relative increase in the amount of oxygen in the
  circulatory system in the region of activated
  brain areas.

5
The BOLD effect - how does it work?See
Logothetis et al(Nature 412(2001), p150)
6
From A Physiology Point Of View
BOLD signal results from a complicated mixture of
these parameters
7

• CBF, CBV, and CMRO2 have different effects on
  HbO2 concentration
• Interaction of these 3 produce BOLD response
• They change Hb which affects magnetic
  environment.

8
Detecting BOLD changes

• This is easy if the BOLD effect is large
  (gt1-2) but not easy when the effects are
  smaller.
• This is because fMRI images are noisy, i.e they
  contain lots of changes in intensity that are NOT
  related to brain activity. These may be caused by
  movement, changes in the electronic or magnetic
  properties of the scanner, heartbeat, breathing
  etc.

9
Timing of BOLD effects
10
Considerations in planning an fMRI experiment in
psychiatry

• What type of experimental design should we use?
• What type of data should we collect ( how many
  slices, how fast should we collect individual
  brain volumes?
• How many subjects do we need?

11
Experimental design

• How many conditions?
• Blocked or event related?

12
How many conditions?

• All fMRI experiments ( almost!) work on contrasts
  between conditions.
• The simplest experiment we might perform would
  have two conditions.
• The difference between these conditions will
  normally be the factor we wish to test.
• Example - to test a visual response to an image
  as a whole, the baseline might be a blank screen
  and the contrasting condition would be the image.

13

• Most experiments are more complex and have more
  than one active condition.
• It is wise to keep the number of experimental
  conditions as small as possible without
  compromising the experiment.

14
Blocked or event related?

• Blocked designs have groups of stimuli presented
  together. Can get good power to detect effects.
  BOLD effect rises to a plateau , stays at this
  level to the end of the block, then declines. BUT
  can get problems with habituation.

15
Event-related designs

• Event related designs started by presenting
  individual stimuli in isolation - (though still
  need to average multiple events of a particular
  type to get enough statistical power)
• Can have randomised presentation order/timing
  etc. Now have rapid event related resigns where
  events of different types can be presented close
  together.

16
Blocked vs. Event-related
17
Possible Advantages of Event-Related Designs

• Flexibility and randomization
• eliminate predictability of block designs
• avoid practice effects
• Post hoc sorting
• (e.g., correct vs. incorrect, aware vs. unaware,
  remembered vs. forgotten items, fast vs. slow
  RTs)
• Can look at novelty and priming
• Rare or unpredictable events can be measured
• e.g., P300
• Can look at temporal dynamics of response
• Dissociation of motion artifacts from activation
• Dissociate components of delay tasks
• Mental chronometry

Source Buckner Braver, 1999
18
Data acquisition?

• How many slices? - ideally want high resolution
  i.e thin slices but very thin slices cause loss
  of signal/noise in data.
• How much of brain should be covered? Usually want
  whole brain but with a STRONG hypothesis might
  use limited coverage at high resolution
• How fast should brain volumes be collected?
  Faster collect, fewer slices, but for good event
  related designs, need short TRs ( fast
  collection)
• Usually, have to compromise between conflicting
  requirements. Need to prioritise.

19
fMRI Experiment Stages Prep

• 1) Prepare subject
• Consent form
• Safety screening
• Instructions
• 2) Shimming
• putting body in magnetic field makes it
  non-uniform
• adjust 3 orthogonal weak magnets to make
  magnetic field as homogenous as possible
• 3) Sagittals
• Take images along the midline to use to plan
  slices

20
fMRI Experiment Stages Anatomicals

• 4) Take anatomical (T1) images
• high-resolution images (e.g., 1x1x2.5 mm)
• 3D data 3 spatial dimensions, sampled at one
  point in time
• 64 anatomical slices takes 5 minutes

21
Slice Terminology
22
How many subjects?

• Ideally need to do a power calculation.
• BUT for power calculation need effect size.
  Usually dont know effect size for new
  experiments.
• ALSO effect size or detection of effects may
  show regional variation. Need power calculation
  for each region.
• GROUP size should be based on region of interest
  with the LEAST power to detect a response or
  response difference.

23
What do real experimental responses look like?
24
(No Transcript)
25

• In many psychiatric fMRI studies, the effect
  sizes are likely to be of this order or less.
• It is important for us to know what sizes of
  effect we can detect.
• For this reason, we have done some detailed
  simulations on groups of subjects to examine how
  response detection and its reliability changes
  with brain region, experimental design and effect
  size.

26
An example

• We have simulated an experiment that activates
  anterior cingulate and bilateral hippocampus in a
  group of 10 men, aged 20-65.
• We scanned them for five minutes, without asking
  them to do an experiment, to obtained baseline or
  null fMRI data.
• For each person, we obtained whole brain fMRI
  data at 3 second intervals throughout the 5
  minutes.

27
Baseline error testing

• The first thing we did was to analyse the data
  with using a variety of experimental designs we
  commonly use, both blocked ( stimuli presented
  together in blocks) or event-related ( stimuli
  presented separately).
• Ideally, under these conditions we should only
  see random errors. The error rate should be that
  predicted by the p value and the number of voxels
  in the brain at which we perform the analysis.

28
Embedded activations

• We then embed our anterior cingulate /
  hippocampal responses in the baseline brain
  images at different response levels and using
  different designs to see how well we pick it up.
• We vary the type of experiment ( event-related,
  blocked) and the number of simuli and spacing of
  stimuli.

29
Simulation 1 - Blocked Experimental Design
30 Seconds
30 Seconds
5 Minutes

30
No embedded activation, expected number of false
positive clusters lt 1.
31
0.25 embedded activation, expected number of
false positive clusters lt 1.
32
0.5 embedded activation, expected number of
false positive clusters lt 1.
33
1 activation, expected number of false positive
clusters lt 1.
34
(No Transcript)
35
Summary of data acquired in simulations

• We have many more results than this using various
  simulated event related designs ( different
  numbers of events, different effects sizes ) and
  various types of block design ( different block
  lengths and effect sizes).

36
Conclusions

• In simulations of a wide variety of experimental
  designs assuming that ALL members of a group of
  10 subjects respond and that there are responses
  to ALL stimuli ( i.e. the best case scenario),
  the limit of detection of the whole network of
  simulated activations lies at a BOLD effect level
  of around 0.25.

37
Conclusions ( continued)

• Even at this level, detection of responses seems
  to vary from region to region ( probably because
  of noise issues ) and design to design ( power
  issues).
• There is evidence that response detection in the
  temporal lobes/ hippocampus/ amygdala is lower
  than in regions such as the anterior
  cingulate/prefrontal cortex.

38
Conclusions continued

• With an detection limit around 0.25 BOLD
  changes in groups of the size often used in
  psychiatric imaging, we should realise that in
  order to detect an increase in an effect size of
  this order, the response would need to double!!
  As anything smaller than 0.25 would fall below
  detectability, 0.20 or 0.15 might appear as
  zero. Decreases in an already small effect are
  probably not going to show up.
• Subtle changes may be hard to detect.

39
Recommendations

• If you need to use event-related designs you may
  compromise power.
• Group sizes gt 10 are likely to be required if
  effects are small.
• If possible, estimate effect size to see what the
  power of your experiment may be. Some estimate
  could be made in your likely regions of interest
  using a few subjects ( say 5 -6).
• This might save investing large amounts of money
  on experiments that are likely to lack the power
  to test your main hypotheses.

40
Final words

• Despite its problems, fMRI has proved to be a
  useful and versatile tool in detecting changes in
  brain function not easily accessible by other
  techniques.
• It can be a powerful tool but healthy scepticism
  is a good approach.
• ALWAYS check your results as early and thoroughly
  as possible.
• Look at the responses you are getting. Do they
  fit hypotheses and previous work?