Maps and metaphors the perils of functional neuroimaging

1
Maps and metaphors the perils of functional
neuroimaging

• Paul Fletcher
• pcf22_at_cam.ac.uk

2
The basis of the functional imaging signal
rest
rest
3
The basis of the functional imaging signal
P.E.T.
Task
4
The basis of the functional imaging signal fMRI
5
Why is brain mapping difficult?
Y(1-n) c ß.X (1-n) ?

• Heterogeneity of macrostructure across
  individuals?
• Heterogeneity of microstructure across
  individuals?
• Heterogeneity of macrostructure-microstructure
  relationship across individuals?
• Assumption of parcellated function not entirely
  predicted by structural observation

6
Implications of macro- and microstructural
patterns

• Group maps of task-related activation are likely
  to produce type II error
• Type I error feasible though unlikely
• Problems addressable through probabilistic brain
  maps
• The search for a neat functional parcellation,
  however, may be misconceived

7
Why might brain mapping be impossible?
Y (1-n) c ß.X (1-n) ?

• The relationship may not exist with Y as measured
  by functional neuroimaging.
• Relationship between X and Y (ß) may be
  context-dependent/non-stationary. Adaptive
  Coding? Equipotentiality?
• Relationship between X and Y may be expressed as
  a change in relationship between Y1 and Y2-x
  i.e. connectivity change.
• X specifies a task/setting that is usually
  formulated in terms of a manipulation of a
  cognitive process. Ultimately, therefore, brain
  mapping is predicated upon a belief that
  cognition is accessible and analysable.

8
Why might brain mapping be impossible?
Y (1-n) c ß.X (1-n) ?

• The relationship may not exist with Y as measured
  by functional neuroimaging.
• Relationship between X and Y (ß) may be
  context-dependent/non-stationary. Adaptive
  Coding? Equipotentiality?
• Relationship between X and Y may be expressed as
  a change in relationship between Y1 and Y2-x
  i.e. connectivity change.
• X specifies a task/setting that is usually
  formulated in terms of a manipulation of a
  cognitive process. Ultimately, therefore, brain
  mapping is predicated upon a belief that
  cognition is accessible and analysable.

9
rest
rest
10
Why might brain mapping be impossible?
Y (1-n) c ß.X (1-n) ?

• The relationship may not exist with Y as measured
  by functional neuroimaging.
• Relationship between X and Y (ß) may be
  context-dependent/non-stationary. Adaptive
  Coding? Equipotentiality?
• Relationship between X and Y may be expressed as
  a change in relationship between Y1 and Y2-x
  i.e. connectivity change.
• X specifies a task/setting that is usually
  formulated in terms of a manipulation of a
  cognitive process. Ultimately, therefore, brain
  mapping is predicated upon a belief that
  cognition is accessible and analysable.

11
Is cognitive function accessible/analysable? (is
brain mapping possible?)

• Are we confident that we can describe brain
  function in terms of a set of cognitive
  processes?
• Are we able to design tasks that will isolate and
  manipulate these sub-processes?
• Do we have sufficient knowledge of process
  inter-relationships to predict accurately
  interactions at the cognitive and functional
  anatomical level?

12
Respond
Process x y z
Process x y z A
Perceive

• I didnt realise how many ants spontaneously
  combust until I started to look at them closely
  through a magnifying glass on bright, sunny days
• Harry Hill

13
Describing cognitive processes

• A description of cognitive functions involves the
  adoption of a high level figurative terminology
• Metaphor
• Monitoring?
• Error checking?
• Strategy selection?
• Metonymy
• a series of working memory studies
• memory encoding tasks

14
The dangers

• Pseudo-insight through terminology
• Jaspers, 1963
• Assumption that a figurative description
  represents a neurobiological reality
• Circularity
• You can tell a lot about someones personality by
  what theyre like
• Harry Hill

15

• With possible exceptions, it appears highly
  unlikely that any single study will unambiguously
  identify a one-to-one mapping of cognitive
  process to brain region because our functional
  architecture is incomplete.
• At present, brain mapping in its most widely
  used sense is inappropriate as an end in itself
• Our initial goal must be to establish, where
  possible, a more complete taxonomy of cognitive
  function.
• How may functional neuroimaging participate in
  this?

16
Functional neuroimaging as part of the
development of a cognitive taxonomy

• 1. To indicate where differing tasks engage same
  processes or where apparently identical tasks
  engage different processes
• meta-analyses
• family resemblances
• 2. Establishing a neurobiological marker for a
  theoretical model
• Linking brain to theory and behaviour

17
1. Same tasks differing processes
Post-scan recognition task.
18
Subsequent recognition success
19
1. Differing tasks same processes
Left IFC and encoding -Predictive of subsequent
retrieval -necessary (?) to subsequent
retrieval -Encoding instructions are incidental
Shallice et al, Nature 1994
20
Activation of left IFG seen across many tasks

• Semantic tasks
• Raichle et al C Cort 1994
• Peterson et al Nature 1988
• Demb et al J Neurosci 1995
• Vandenberghe et al, Nature 1996
• Binder et al J Neurosci, 1997
• Gabrieli et al PNAS 1998
• Poldrack et al NeuroImage 1999
• Wiggs et al Neuropsychologia 1999

• Encoding tasks
• Kapur et al, PNAS 1994
• Shallice et al, Nature 1994
• Haxby et al,PNAS 1996
• Kelley et al, Neuron 1998
• Wagner et al, Science 1998
• Kopelman et al, Brain 1998
• Fletcher et al, Brain 1998
• Otten et al, Brain 2000

21
Episodic Encoding and semantic tasks family
resemblance?
Event-related fMRI study of paired associate
encoding 120 items (word associate pairs) -60
closely related KingQueen DogCat -60
distantly related BoatNet PrinceSkull F
letcher, Shallice Dolan NeuroImage, 2000
22
Distant versus close pairings
23
L IFC effects Distant vs Close pairs
24
Episodic encoding and semantic processing are
therefore associated with left IFG. Can we
dissociate these two sets of processes in
neuronal terms?
25

• A proactive interference task may engage semantic
  processes while suppressing episodic encoding
• An increased left IFG response in the face of
  proactive interference therefore links this
  region more strongly to semantic processing

26
Pre-scan
Scan
List 3 Old-Old GameBridge DogBoxer New-New Ston
eGranite ClothVelvet Old-New GameFootball Dog
Dalmation NewOld StructureBridge AthleteBoxer
List 1 GameBridge DogBoxer
List 2 GameBridge DogBoxer
27
Proactive interference (PET study)- Dolan
Fletcher, Nature 1997
28
Proactive interference fMRI study Fletcher,
Shallice Dolan NeuroImage, 2000

• LabourBaby
• ChickBird
• NotesStudy
• PartyBallroom
• X 4

• PartyLabour
• BirdNotes
• ChickBaby
• BallroomStudy

29
Increasing familiarity across first four
repetitions
30
Rearranged versus novel pairs
31
L IFG an emerging pattern?

• Tulving et al - Retrieval of semantic information
• Gabrieli et al - Maintenance of semantic
  information
• Thompson-Schill et al- Selection from among
  competing semantic attributes
• Wagner et al- Control of semantic retrieval
• Ultimately, one intriguing thing to emerge across
  the family of tasks showing left IFG activation
  is that they may be explained without recourse to
  episodic memory encoding

32
2. A neurobiological marker for a theoretical
model Associative learning (Fletcher et al,
Nature Neuroscience 2001)
Advantage existence of a theoretical model that
adequately reflects human behaviour. Does this
model have a neurobiological plausibility? Can
we use the specificity of this model to explore
frontal function in a way that minimises the
uncertainties arising from figurative language
and inexact process specification?
?V a ß (? SV)
33
?V a ß (? SV)
Unpredictability Learning
34
?V a ß (? SV)
Learning modulated by configuration of
unpredictable event
35
?V a ß (? SV)
36
(No Transcript)
37
(No Transcript)
38
Brain
Behaviour
Theoretical Model
?V a ß (? SV)
Subsequent Predictive responses
39
In brief

• Mapping of cognitive processes onto brain
  structure is rendered ambiguous by
• Heterogeneity of micro- and macro-structure
• An incomplete knowledge of functional
  architecture

40
Functional neuroimaging may contribute to the
development of a functional architecture -

• Through identification of neurobiological
  similarities across apparently different tasks
• Through identification of neurobiological
  differences across apparently similar tasks
• Neuroimaging as an extra-sensitive behavioural
  measure?
• e.g. Dm effects
• Through establishment of the plausibility of
  cognitive models of behaviour at the level of
  brain function

41
Future for functional neuroimaging

• Psychopharmacological manipulations
• Measures of functional integration

42
Acknowledgements
London R Dolan K Friston C Frith R Frackowiak R
Henson M Rugg T Shallice W.D.C.N
Perth R Honey P Michie
Cambridge J Anderson E Bullmore Caroline
Stephenson W.B.I.C
Düsseldorf G Fink N Palomero-Gallagher K
Zilles I.M.E.