BOLD fMRI

1
BOLD fMRI John VanMeter, Ph.D. Center for
Functional and Molecular Imaging Georgetown
University Medical Center
2
Outline

• BOLD contrast fMRI conceptually
• Relationship between BOLD contrast and
  hemodynamics
• History of BOLD contrast
• Relationship between neuronal glucose metabolism
  and blood flow
• Theories and properties of BOLD contrast
  mechanisms

3
Neuronal Activity and Blood Flow Changes Initial
Hypothesis

• Roy and Sherrington hypothesized that local
  neuronal activity is related to regional changes
  in both cerebral blood flow and metabolism
  (1890).
• There are, then, two more or less distinct
  mechanisms for controlling the cerebral
  circulation, viz. - firstly, an intrinsic one by
  which the blood supply of the various parts of
  the brain can be varied locally in accordance
  with local requirements, and secondly, an
  extrinsic, viz. - the vasomotor nervous system

4
Roy and Sherringtons Experiments

• the increase in the volume of the brain which
  results from stimulation of the sensory nerves is
  mainly if not entirely due to passive or elastic
  distension of its vessels as a result of the
  blood-pressure in the systemic arteries.

5
History of BOLD fMRI

• Initial discovery of magnetic properties of blood
  by Linus Pauling and graduate student Charles
  Coryell (1936)
• Magnetic properties of a blood cell (hemoglobin)
  depends on whether it has an oxygen molecule
• With oxygen ? zero magnetic moment
• Without oxygen ? sizeable magnetic moment

6
Initial In Vivo Measurement of Neuronal Activity

• Initial techniques used PET (positron emission
  tomography)
• PET uses injection of a radiotracers which are
  variants of physiological molecules that include
  a radio isotope
• FDG (2-fluoro-2deoxy-D-glucose) for glucose
  metabolism
• H2015 for blood flow

7
Functional Imaging - PET

• Sokoloff demonstrated that rCBF (blood flow)
  increases in visual cortex in proportion to
  photic stimulation using PET (1961).
• Demonstrated coupling between blood flow and
  metabolism (1981).

8
Relationship Between Glucose Metabolism and Blood
Flow

• Sokoloff (1981) used autoradiography
• Measured both glucose metabolism and blood flow
• 39 brain regions in rat brain
• Correlation r0.95
• Slope m2.6

9
First MRI-based Measurement of Neuronal Activity

• Belliveau (1990) used MRI contrast agent
  Gadolinium as an exogenous tracer
• Gadolinium locally disrupts MRI signal
• Perfusion weighted imaging (PWI)

10
Oxy- vs. Deoxy- Hemoglobin

• Oxygenated hemoglobin (Hb) is diamagnetic (zero
  magnetic moment)
• Deoxygenated hemoglobin (dHb) is paramagnetic
  (magnetic moment)
• Magnetic susceptibility of dHb is about 20
  greater than Hb
• Magnetic susceptibility affects rate of dephasing
  - T2 and T2 contrast!

11
T1 T2 Contrast Versus Oxygenated Hemoglobin
12
Demonstration of BOLD Contrast

• Seiji Ogawa (1990) manipulates oxygen content of
  air breathed by rats
• Results in variation of oxygenated state of blood
  
• Demonstrates effect on T2 contrast to make
  images of blood vessels

13
Ogawas Images of Blood Vessels Based on Oxygen
Content

• Pure oxygen
• Normal Air

14
Magnetic Susceptibility Greater on T2 than T2
Images
Spin Gradient Echo (T2) Echo (T2)

• Oxygenated
• Hemoglobin
• Deoxygenated
• Hemoglobin

15
Oxygenation vs Local Field Changes
Bandettini and Wong. Int. J. Imaging Systems and
Technology. 6133 (1995)
16
First fMRI BOLD in Human

• Kwong (1992) demonstrated first BOLD-contrast
  fMRI in human visual cortex

17
Blood Flow vs BOLD Changes

• Kwong also showed how changes in BOLD
  corresponded to changes in blood flow
• Important to show that BOLD and blood are related

18
Build Up to BOLD Contrast

• Hypothesis of relationship between blood flow and
  activity (Roy Sherrington, 1890)
• Discovery of differential magnetic properties of
  oxygenated and deoxygenated hemoglobin (Pauling,
  1936)
• Blood flow increases with activity (Sokoloff,
  1961)
• Blood flow correlated with glucose metabolism
  (Sokoloff, 1981)
• Demonstration of blood flow measured using MRI
  with an exogenous tracer (Belliveau, 1990)
• Demonstration of effect of dHb on T2 contrast
  (Ogawa, 1990) use of blood as an endogenous
  tracer
• Generation of first BOLD images (Ogawa, 1990)
• First BOLD images in humans (Kwong, 1992)

19
Basic Model of Relationship Between BOLD fMRI
Neuronal Activity
20
Disparity Between Blood Flow Oxygen Consumption

• Fox Raichle conducted PET experiments to
  measure glucose metabolism (CMRglu), blood flow
  (CBF), and rate of oxygen metabolism (CMRO2)
• Measured percent change between visual
  stimulation and rest
• Increase in CBF50, CMRglu51
• But increase in CMRO2 is only 5!!
• Implies anaerobic metabolism of glucose

21
(No Transcript)
22
Disparity MRI Signal Increase

• Upshot of Fox Raichle much more oxygen (CBF)
  is supplied than is used (CMRO2)
• While neuronal activity results in more
  deoxygenated hemoglobin much more oxygenated
  hemoglobin flows in flushing out deoxygenated
  hemoglobin
• Result is a decrease in dHB and thus an increase
  in MRI signal
• But theres uncoupling of glucose metabolism and
  oxygen metabolism - WHY?

23
Uncoupling Problematic

• Fox Raichle data nicely explains why MRI signal
  increases with neuronal activity
• But a new problem is presented uncoupling of
  glucose and oxygen metabolism
• We expect a 61 ratio of oxygen-to-glucose (OGI)
  for aerobic glycolysis but FR saw about 110
• Implication is anaerobic glycolysis is used

24
Theories to Explain Uncoupling Found by Fox
Raichle

1. Watering the Garden for the Sake of One Thirsty
   Flower
2. Astrocyte-Neuron Lactate Shuttle Model
3. Transit Time and Oxygen Extraction

25
Separate Measurement of Oxy Deoxy Hemoglobin

• Malonek Grinvald used optical imaging to
  measure Hb and dHb separately during visual
  stimulation
• ?dHb spatially focal and co-located to neuronal
  activity
• ?Hb more widely distributed

26
Implications of Differences in Concentration of
Hb dHb

• Rapid increase in dHb implies oxidative
  metabolism initially
• High spatial correspondence between initial dHb
  increase and neuronal activity
• Coarse spatial correspondence and greater extent
  of delivery of Hb

27
Theories to Explain Uncoupling Found by Fox
Raichle

1. Watering the Garden for the Sake of One Thirsty
   Flower
2. Astrocyte-Neuron Lactate Shuttle Model
3. Transit Time and Oxygen Extraction (extended to
   Balloon Model)
4. Aerobic glycolysis already near max at rest thus
   activity requires quick increase in energy via
   anaerobic glycolysis (Prichard, 1991)

28
Watering the Garden

• According to this model uncoupling observed by
  Fox Raichle does not imply anaerobic glycolysis
• Instead Malonek Grinvalds data shows huge
  excess of freshly oxygenated hemoglobin spread
  over a wide area displacing deoxygenated
  hemoglobin
• But CMRglu wasnt measured still havent
  explained why Fox Raichle gets a 110 versus
  expected 61 OGI

29
Astrocyte-Neuron Lactate Shuttle Model

• Initially anaerobic glycolysis occurs producing
  excess glutamate (consistent with Fox Raichle)
• Glutamate taken up by astrocyte to prevent
  toxicity and converted to glutamine which neuron
  can reuse
• Delicate balance is achieved by astrocyte through
  intake of Na produced by sodium-potassium pump
  of neuron
• Astrocyte uses 2 ATP molecules
• Great because thats all the ATP available!
• But wheres the ATP for the neuron?

30
ANLS Model (contd)

• Astrocyte dumps resulting lactate, which diffuses
  into neuron that turns into pyruvate and into TCA
  cycle to give neuron 36 ATP molecules for
  neurons energy
• Thus, were back to aerobic glycolysis, which
  requires 6 molecules of oxygen
• Model hypothesizes early anaerobic followed by
  aerobic glycolysis
• Support for this comes from Mintun (2002) who
  showed uncoupling only occurs with initial onset
  of stimulus then coupling is reestablished with
  continued stimulation

31
Astrocyte-Neuron Lactate Shuttle Model
32
Transit Time and Oxygen Extraction

• Disputes that uncoupling implies anaerobic
  glycolysis as does Watering the Garden
• Model is based on limited time for extraction of
  oxygen due to increase in velocity of blood flow
  with neuronal activity

33
Transit Time and Oxygen Extraction

• Model proposed by Buxton (1998) rests on four
  assumptions
• Increased blood flow accomplished by increase in
  velocity as opposed pumping more blood through
  more capillaries
• Virtually all oxygen is metabolized
• But not all of the glucose is metabolized
• Extraction of oxygen from blood by neurons is
  limited and proportional to transit time
• Transit time - how long it takes for blood to
  pass through a given area

34
Transit Time and Oxygen Extraction

• Wouldnt limited time for extraction of oxygen
  due to increase in velocity of blood also limit
  glucose availability?
• Buxton - well actually glucose availability is
  even more limited than oxygen but less than half
  that is extracted is actually used
• Data from Gjedde (2002) supports glucose part

35
Balloon Model

• No uncoupling of CBF and CMRO2 difference
  between CBF and CMRO2 lowers oxygen extraction
  fraction (E) Fick Principle
• Initial increase in blood flow increases blood
  volume (ballooning of venous capillary to
  accommodate)
• Predicts an initial dip in BOLD signal

Buxton et al. Neuroimage 2004
36
Uncoupling Problem

• Debate continues to this day
• Uncoupling problem important to understanding the
  fundamental basis of fMRI signal
• fMRI is an indirect measure of blood flow and is
  not directly tied to glucose metabolism or even
  oxygen metabolism
• Relationship between mechanisms of metabolism and
  blood flow is important to understanding how
  closely related BOLD and blood flow are to
  neuronal activity

37
Implications of Theories for Uncoupling

• Watering the Garden model posits widespread
  distribution of CBF increase ? poor fMRI spatial
  resolution
• Transit Time model implies excess oxygen rich
  blood passing over area of activity getting into
  venous system ? poor fMRI spatial resolution
• Both imply a Draining Vein problem with dHb
  flowing downstream of area of activity
• Frahm (1994) asked Brain or Vein?
• Uncoupling issue remains unresolved

38
Physiological Mechanisms for Regulation of Blood
Flow

• How is blood flow controlled?
• Arterioles well upstream need to respond to
  produce local changes in blood flow
• Mechanism for accomplishing this is largely
  unknown
• Possible candidates include nitrous oxide
  synthesis, potassium accumulation, generation of
  lactate, or acetylcholine activity

39
Initial Dip

• Studies used very short TR (100ms) and visual
  stimulus for 10s at 4T or higher
• Menon (1995) found Initial Dip in fMRI signal
  before expected increase
• Theres also a post stimulus undershoot

40
Spatial Extent of Initial Dip

• Voxels with initial dip were more spatially
  restricted and localized to gray matter around
  calcarine sulcus

41
Implications of Initial Dip

• Menon suggested dip is directly related to oxygen
  extraction and thus more closely related to
  neuronal activity
• But dip could also result from temporary decrease
  in blood flow or increase in blood volume
• Initial dip if it occurs is contradictory with
  anaerobic glycolysis - Why?
• Balloon model predicts increase in blood volume
  and thus consistent with initial dip but for a
  different reason than Menon posits

42
HDR (Hemodynamic Response)HRF (Hemodynamic
Response Function)

• Change in MR signal related to neuronal activity
  (HRF)
• Has multiple components
• Changes delayed by 1-2 sec (lags activity)
• Initial dip (not always seen)
• Influx of Hb greater than needed for activity
• 5-6 sec time to peak
• Undershoot follows 6s after peak

43
Typical HDR for Long Stimulus (Block)

• Peak is sustained with prolonged stimulation
• Block is also referred to as an epoch
• Brief stimulus is referred to as an event

44
Undershoot

• Arises from rapid return to baseline of CBF but
  delayed return of CBV
• Delay in CBV return to baseline results in an
  accumulation of dHb

45
BOLD vs Neuronal Activity

• Logothetis, et al., 2001 recorded LFP, MUA, SUA,
  and BOLD simultaneously
• BOLD response best explained by changes in LFP
• Suggests BOLD reflects incoming input and local
  processing rather than spiking activity
• The BOLD contrast mechanism directly directly
  reflects the neural responses elicited by a
  stimulus.

46
Open Questions about Basis of BOLD fMRI

• Uncoupling problem - Why does it occur? To what
  extent?
• Is there an Initial Dip? What causes the dip? Is
  it more localized than the expected signal
  increase?
• What about Draining Veins?
• How does the arterial system upstream know when
  and by how much to increase blood flow?

47
Factors Affecting BOLD Signal

• Physiology
• Cerebral blood flow (baseline and change)
• Metabolic oxygen consumption
• Cerebral blood volume
• Equipment
• Static field strength
• Field homogeneity (e.g. shim dependent T2)
• Pulse sequence
• Gradient vs spin echo
• Echo time, repeat time, flip angle
• Resolution

48
Physiological Baseline

• Baseline CBF changes (up for hypercapnia, down
  for hypocapnia)
• But ?CBF ?CMRO2 unchanged (probably) (Brown et al
  JCBFM 2003)
• BOLD response ? (probably)

Cohen et al JCBFM 2002
49
Spatial Temporal Properties of BOLD

• Spatial resolution - ability to distinguish
  unique changes in activity from one location to
  the next
• Temporal resolution - ability to distinguish
  changes across time
• Linearity vs Nonlinearity - does combined
  response to 2 or more events with short ISI
  (inter-stimulus interval) lead to sum in BOLD
  response?

50
Image Resolution (2D)

• FOV - Field of View, prescribed area that will be
  covered in the acquisition
• Matrix size - how many voxels will be acquired in
  each dimension
• Rectangular FOV possible
• Voxel dimension (size)
• FOV/matrix

51
Example

• FOV 192mm x 192mm
• Matrix 64x64
• What is the voxel size in-plane?
• 3mm x 3mm

52
Slice Thickness Defines 3rd Dimension

• Does not have to match size of in-plane
  resolution
• Voxels are referred to as isotropic when all
  three sides have the same size
• Gaps between slices can be used to cover more of
  the brain
• 3D Acquisition has a 2nd phase encode for through
  plane dimension and effectively 3rd FOV dimension
  but usually presented on console as slice
  thickness

53
Problems With Increasing Spatial Resolution

• Increased spatial resolution results in smaller
  voxels
• Fewer protons so less MRI signal
• Less dHb thus more noise in BOLD fMRI signal
• Degree of activation varies by brain region with
  greater activation in sensorimotor areas and less
  in frontal and association cortices
• Smaller voxels ultimately make detecting changes
  harder

54
Spatial vs Temporal Resolution

• Acquisition time per slice goes up as voxel size
  goes down
• Number of phase encode lines increases thus more
  time required to cover k-space
• Decreasing slice thickness will require
  increasing number of slices to maintain same
  coverage again increasing acquisition time

55
Designing an fMRI Protocol

• Tradeoffs
• Increased spatial resolution requires
• Increased TR (scan time)
• Less coverage (fewer slices)
• Increased temporal resolution requires
• Decreased spatial resolution (larger voxels)
• Less coverage (fewer slices)
• Reducing amount of k-space acquired (less SNR)
• Increased SNR (signal-to-noise ration) requires
• Decreased spatial resolution and/or
• Increased scan time via averaging

56
(f)MRI Image Acquisition Constraints
Signal to Noise Ratio
Spatial Resolution
Temporal Resolution
57
Partial Volume Effects

• Any given voxel will be a mix of tissue types
• Boundaries with sulci will include CSF
• Both can lead to a reduction in overall fMRI BOLD
  signal

58
Spatial Correspondence
59
Theoretical Lower Bound on Spatial Resolution

• Ultimately determined by the size of capillaries
• 1mm in length
• 100 microns between capillaries
• Theoretical lower bound for any hemodynamic based
  measurement is 100 microns

60
Mapping Ocular Dominance Columns

• Menon, 1997 presented visual stimulus to
  alternating eyes
• Expect to see side-by-side alternating areas of
  activation in V1 corresponding to columns first
  shown by Hubel Wiesel
• Acquired at 4T using a single slice with 547?m x
  547?m resolution

61
fMRI of Ocular Dominance Columns
62
Ocular Dominance Columns - Take 2

• Cheng, 2001 used 4T with 470?m2 resolution,
  single slice
• Each slice required 32-RF pulses to get enough
  SNR (averaging), scan time for 1 slice was 10s!
• Stimulus was 2min monocular presentation of light
  interspersed with 1min darkness

63
Replication Within Subject
64
Ocular Dominance Columns - Take 3
65
fMRI Data Processing Spatial Resolution

• Typical processing includes
• Motion correction which will reslice the data
  (reslicing of data requires averaging of voxels
  to reformat data)
• Spatial Normalization (transforming into atlas
  space) again reslices data
• Spatial smoothing (blurring)
• Net result is reduction in effective spatial
  resolution

66
Temporal Resolution

• TR in fMRI refers to time needed to collect one
  volume of data
• Long TR (gt3s) good for detecting differences in
  activation but not differences in HRF
  (hemodynamic response function) characteristics
• Where is activity occurring?
• Shorter TR (lt2s) gives better estimate of
  differences in HRF characteristics
• What are the differences in activity between two
  stimuli activating in the same area?

67
(No Transcript)
68
JitterInterleaved Stimulus Presentation

• Instead of locking stimulus presentation to the
  TR jitter it
• Effectively gives more data on HRF curve than
  locked to the TR
• Thus, effective temporal resolution is increased

69
BOLD as a Psychophysical Measure
70
Duration of Cognitive Processing BOLD Response

• Psychophysical experiments looking at mental
  rotation have shown that the greater the
  differences in angle between two figures the
  longer the response time
• What happens to BOLD response?

71
BOLD Response Duration Increases
72
Timing Between Brain Regions

• Move joystick from one target to another
• Measured reaction time and difference in onset
  time of BOLD response different brain regions
• V1-SMA differences suggests decision pathway
• SMA-M1 flatness suggests simple execution

73
Latency of BOLD Response

• Examination of the latency (time to onset) in
  voxels with significant activation
• Blue shortest
• Yellow longest
• Output from V1 (slices a c) feeds fusiform
  gyrus (slices b d)
• As hoped response delayed in fusiform relative to
  V1

74
Linearity of Hemodynamic Response?

• Linearity would imply
• there is an additive effect of two stimuli
  presented close enough in time
• HRF scales with stimulus intensity
• HRF response to two or more stimuli equal
  summation of response to individual stimuli
• Under what conditions is HRF linear?

75
Linearity of HRF - Theoretical

• Give two stimuli close in time
• Is the HRF for the second equal to the first?

76
Nonlinearity Via Attenuation - Theoretical

• Or is there some attenuation (reduction) in the
  response to the 2nd stimulus?
• Refractory effects - change in response to 2nd
  stimulus based on presence of first?

77
Does HRF Scale with Stimulus Magnitude?
78
Superposition of HRF ?
79
Evidence for Linearity

• Boynton, 1996
• Presented several short stimuli for various
  durations
• Found response scaled with contrast
• Found good correspondence between actual response
  and predicted thus linearity held

80
Superposition

• Boynton found good correspondence between
  predicted and actual measured response
• However, when 2 or more 3s stimuli presented -
  got smaller than predicted response
• Attributed to adaptation of neurons leading to
  reduced activity
• Support for linearity superposition when
  stimuli gt3s

81
Response to Multiple Trials

• Dale Buckner, 1997
• Three identical trials presented
• ISI was either 2s or 5s
• Each trial gives additive effect

82
Separation of Response to Multiple Trials

• Recovered HRF for 2nd and 3rd trials quite
  closely match that of the 1st for 5s ISI
• Again at shorter ISIs of 2s results were reduced
  amplitude and greater latency
• Evidence of nonlinearity at short ISIs

83
HRF as a Function of Interstimulus Interval

• Huettel, 2000 used visual stimuli separated by a
  variable amount of time
• Found reduction in amplitude of response and
  increase in latency as ISI decreased

84
Linearity of HRF and Refractory Period

• Linearity seems to hold for combinations of
  stimuli with ISIs 5-6s or longer
• Much evidence of a refractory period during which
  additional presentation of stimuli produces
  smaller and delayed response
• Is this bad? Can we take advantage of this?

85
fMRI Adaptation (fMRI-A)

• Grill-Spector Mallach, 2001
• Presented same face with different sizes,
  positions, shading, and angles
• Response in fusiform was reduced during
  conditions where size and position was varied
• Signal recovered when shading or angle was
  varied!
• Conclusion - fusiform recognizes identity
  regardless of size or position but treats shading
  and angle changes as different face

86
(No Transcript)
87
fMRI Adaptation

• Top graph - release of response to attributes
  other than color thus this area preferentially
  responds to changes in physical characteristics
• Bottom graph - release of response only to
  vehicle type thus this area preferentially
  responds to complex object categories

88
Summary

• fMRI BOLD signal arises from changes in
  oxygenated state of blood
• Blood flow is primary means for delivering oxygen
  and glucose to neurons for production of energy
• Aerobic and anaerobic glycolysis implies
  different amounts of ATP (energy) production and
  oxygen requirements important for understanding
  how well BOLD relates to neuronal activity
• Definitive linkage of BOLD, blood flow and
  neuronal energy metabolism still elusive
• Properties of BOLD signal