Regions or Blobs

1
Regions or Blobs?
2
Methodological Fundamentalism
The latest review I received
3
The Debate in Print
4
Friston et al.s Critiques and Rebuttals

• ROIs may be context-specific and heterogeneous
• probably more true for some ROIs (e.g., LOC) than
  others (e.g., V1)
• factorial designs are the best way to go
• factorial designs (esp. gt 3-way) are complicated
  to understand
• one may not be interested in all contrasts and
  interactions
• other designs are also valuable (e.g.,
  parametric, adaptation)
• using a contrast in a factorial design as a
  localizer can bias results for other contrasts
• e.g., find area by A gt B
• even in random data set, can find significance
  for A gt C
• but can still do comparisons between even and odd
  runs
• localizers can waste time
• sometimes true but not necessarily true
• multi-session functional brain banks are a
  solution
• caveat frequent subjects may give poorer
  activation
• ROIs preclude any inferences about functional
  specialization
• no one says you cant have multiple ROIs or do
  voxelwise analyses

5
Saxe et al.s Rebuttal

• functional regions can vary considerably in
  stereotaxic space
• ROIs require less stats voodoo
• simple t at p lt .05 often does the trick
• less concern about problems like multiple
  comparisons, serial correlations
• ROI analyses are inherently random effects but
  require fewer subjects than voxelwise approaches
  (because each subject contributes)
• more powerful
• particularly important when looking for subtle
  differences
• intersubject averaging waters down effects
• naming areas should not be taboo

6
Who Uses ROIs?

• ROI approaches are more popular
• in disciplines where specific areas are expected
  from other methods
• e.g., vision science -- rich literature on V1,
  V2, MT, etc.
• with fMRI (vs. PET)
• especially for high-resolution imaging
• for paradigms with subtle differences
• e.g., adaptation
• among users of software that make it easy to use

7
The Crux of the Debate

• Both approaches solve problems of combining data
  across multiple subject and performing random
  effects
• Voxelwise Approaches
• cover the whole volume
• require no assumptions
• advocate smoothing
• require averaging in a common space
• require rigourous statistics
• random effects
• correction for multiple comparisons
• can do small volume correction
• correction for serial correlation
• ROI Approaches
• are limited to a subset of areas
• require assumptions
• maintain high resolution
• can account for anatomical variability
• require only simple stats

8
My Assumptions

• The brain contains subregions with distinct
  anatomical features and functional specificity
• These regions may be in consistent locations with
  respect to sulci, but sulci can be highly variable

Watson et al., 1995
Tamraz Comair
9
Sample Region 1 AIPGrasping gt Reaching
Localizer(Kroliczak et al., Neuroscience 2006)
10
Sample Region 2 LOC
LOC contains at least two distinct functional
regions, possibly several Treating LOC as a
single ROI may be problematic
11
Consistency of Sulcal Landmarks

• AIP is consistently at the junction of the
  Postcentral and Intraparietal Sulci
• Unfortunately, this junction is highly variable
  in stereotaxic space

12
Effects of Adjacent Areas
S1
Data 4 x
AIP
Single subject ROI Analysis
13
Effects of Adjacent Areas
S1
Data 4 x
AIP
Multi-subject ROI Analysis
BSC
A
B
Condition
14
Exclusion of Adjacent Areas
Voxelwise Analysis on Smoothed Data
15
Future DirectionsBetter Normalization

• e.g., MNI template, cortex based alignment in
  spherical space
• areas of high anatomical variability (esp.
  parietal cortex) may still be problematic

A
CentralS
Purple overlap of sulci between two
subjects Orange and Blue sulci present in only
one subject
Medial
PostCS
Lateral
IPS
Note that the parietal cortex typically has
the worst overlap
P
16
Future DirectionsHigh-Resolution Imaging
raw data, n1 1.5 x 1.5 x 2 mm interpolated to
1x1x1
spatially smoothed data Gaussian FWHM 8 mm
vision action
action only
17
Future DirectionsCytoarchitectonic Mapping

• may enable probabilistic atlases
• doesnt help the problem of distinguishing
  processing in adjacent areas
• may help constrain definition of functional
  units

hIP1 and hIP2 40 probability maps
18
Comparing the two approaches

• Voxelwise Analyses
• Requires no prior hypotheses about areas involved
  
• Includes entire brain
• Often neglects individual differences
• Can lose spatial resolution with intersubject
  averaging
• Can encourage meaningless laundry lists of
  areas that are difficult to interpret
• You have to be fairly stats-savvy and include all
  the appropriate statistical corrections to be
  certain your activation is really significant

• Region of Interest (ROI) Analyses
• Extraction of ROI data can be subjected to simple
  stats (no need for multiple comparisons,
  autocorrelation or random effects corrections)
• Gives you more statistical power (e.g., p lt .05)
• Often requires fewer subjects
• Hypothesis-driven
• Useful when hypotheses are motivated by other
  techniques (e.g., electrophysiology) in specific
  brain regions
• ROI is not smeared due to intersubject averaging
• Easy to analyze and interpret
• Neglects other areas that may play a fundamental
  role
• Works better for well-established areas with
  clear criteria
• If multiple ROIs need to be considered, you can
  spend a lot of scan time collecting localizer
  data (thus limiting the time available for
  experimental runs)

19
A Proposed Resolution

• There is no reason not to do BOTH ROI analyses
  and voxelwise analyses
• ROI analyses for well-defined key regions
• Voxelwise analyses to see if other regions are
  also involved
• Ideally, the conclusions will not differ
• If the conclusions do differ, there may be
  sensible reasons
• Effect in ROI but not voxelwise
• perhaps region is highly variable in stereotaxic
  location between subjects
• perhaps voxelwise approach is not powerful enough
• Effect in voxelwise but not ROI
• perhaps ROI is not homogenous or is
  context-specific

20
Extra Slides
21
The Danger of Voxelwise Approaches

• This is one of two tables from a paper
• Some papers publish tables of activation two
  pages long
• How can anyone make sense of so many areas?

Source Decety et al., 1994, Nature
22
Example The Danger of ROI Approaches
Source Sunaert et al., 1999, Exp. Brain Res.

• At least 17 areas of the brain respond to visual
  motion!
• Why is MT considered THE motion area? Why do
  90 of fMRI experiments on motion only look at
  MT? What are these other areas doing?