1. Functional MRI Daniel Bulte Centre for Functional Magnetic Resonance Imaging of the Brain University of Oxford

2. Overview  BOLD Contrast  Metabolic and cerebral blood flow response  Mechanism of MR signal change  Neurovascular coupling  Noise  Factors affecting BOLD  More detail  Changing physiological baseline  Metabolic modelling

3. “Activation”

4. Deoxy-Haemoglobin paramagnetic different to tissue  =0.08ppm

5. Oxy-Haemoglobin diamagnetic same as tissue

6. Field homogeneity & oxygenation state  Red blood cell  6  m diameter, 1-2  m thick  Susceptibility  An object with differing magnetic properties distorts the field

7. Water  Freely diffusing water is the source of image signal  Two water spaces  Intravascular (blood)  Capillaries and venules  Extravascular - a larger pool  In 50ms (FMRI TE) water diffuses 4 capillary diameters

8. Magnetic field in a vessel B0B0 r   a Inside Cylinder '  = 2  (1-Y)  ' ∆  cos 2 

9. Magnetic field around a vessel B0B0 r   a Inside Cylinder '  = 2  (1-Y)  ' Outside Cylinder ∆  cos 2   r,   sin 2  a/r  2 cos 

10. Vessel orientation  Field inside and outside depends on angle  with respect to B0 Bandettini and Wong. Int. J. Imaging Systems and Technology. 6:133 (1995)

11. Blood oxygenation  Field inside and outside depends on Y, oxygenation Bandettini and Wong. Int. J. Imaging Systems and Technology. 6:133 (1995)

12. Signal dependence  Macroscopic behaviour of NMR, gradient echo signal  More extravascular at high field  BOLD signal depends on the amount of dHb in the voxel  R2* = 4.3  (1-Y) B 0 CBV  R2* = 0.04 {  (1-Y)} 2 B 0 2 CBV (venules, larger vessels) (smaller capillaries)

13. BOLD signal  B lood O xygen L evel D ependent signal  T2* change from the haemodynamic perturbation associated with neural activation

14. Neural activitySignallingVascular response Vascular tone (reactivity) Autoregulation Metabolic signalling BOLD signal glia arteriole venule B 0 field Synaptic signalling Blood flow, oxygenation and volume From neural activity to BOLD signal

15. 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

16. Deoxy- haemoglobin Deoxy- haemoglobin Oxy- haemoglobin Oxy- haemoglobin Neural Activity Oxygen Consumption Image T2* weighted signal Deoxy- haemoglobin Deoxy- haemoglobin Oxy- haemoglobin Oxy- haemoglobin Neural Activity Oxygen Consumption Image T2* weighted signal Deoxy- haemoglobin Deoxy- haemoglobin Oxy- haemoglobin Oxy- haemoglobin Neural Activity Oxygen Consumption Image T2* weighted signal Deoxy- haemoglobin Deoxy- haemoglobin Oxy- haemoglobin Oxy- haemoglobin Neural Activity Oxygen Consumption Image T2* weighted signal Deoxy- haemoglobin Deoxy- haemoglobin Oxy- haemoglobin Oxy- haemoglobin Neural Activity Oxygen Consumption Image Blood Flow Blood Volume T2* weighted signal Deoxy- haemoglobin Deoxy- haemoglobin Oxy- haemoglobin Oxy- haemoglobin Neural Activity Oxygen Consumption Image Blood Flow Blood Volume T2* weighted signal Deoxy- haemoglobin Deoxy- haemoglobin Oxy- haemoglobin Oxy- haemoglobin Neural Activity Oxygen Consumption Image Blood Flow Blood Volume T2* weighted signal Deoxy- haemoglobin Deoxy- haemoglobin Oxy- haemoglobin Oxy- haemoglobin Neural Activity Oxygen Consumption Image Blood Flow Blood Volume T2* weighted signal Deoxy- haemoglobin Deoxy- haemoglobinOxy-haemoglobin Neural Activity Oxygen Consumption Image Blood Flow Blood Volume T2* weighted signal Deoxy- haemoglobin Deoxy- haemoglobin Oxy- haemoglobin Oxy- haemoglobin Neural Activity Oxygen Consumption Image Blood Flow Blood Volume T2* weighted signal Deoxy- haemoglobin Deoxy- haemoglobin Oxy- haemoglobin Oxy- haemoglobin Neural Activity Oxygen Consumption Image Blood Flow Blood Volume T2* weighted signal

17. Haemodynamic changes underlying BOLD time BOLD response, % initial dip positive BOLD response post stimulus undershoot overshoot 1 2 3 0 stimulus CBF CBV CMRO 2 stimulus

18. BOLD contrast  Transverse relaxation  Described by a time constant  Time for NMR signal to decay  Loss of spins phase coherence (out of step)  Spin echo, T2  Time varying field seen by diffusing spins  Gradient echo, T2*  Time varying field seen by diffusing spins  … plus spatial field variation across voxel  Why is magnetic field non uniform?

19. Modelling of the BOLD effect  Effects of oxygenation on T2*  Ogawa et al., J. Biophys., 64:803-812 (1993)  Kennan et al., MRM, 31:9-21 (1994)  Boxerman et al., MRM, 34:4-10 (1995)  Flow and oxygenation coupling  Buxton and Frank, JCBFM, 17:64-72 (1997)  CBV effects  Buxton et al., MRM, 39:855-864 (1998)  Mandeville et al., JCBFM, 19:679-689 (1999)

20. Signal evolution  Monte Carlo simulation  Signal dephasing in the vascular tree amongst vessels of differing size, oxygenation and orientation  Boxerman J. et al. MRM 1995  Deoxy-Hb contribution to relaxation  Gradient echo S = S max. e -TE/R2*  R2*  (1-Y)  CBV Y=O 2 saturation b~1.5

21. Echo time and BOLD sensitivity BOLD contrast-to-noise optimised when TE~T2* T2* shorter at higher field TE (ms) Relative CNR TE optimization similar to T2 structural optimizations, just between different states rather than different tissues

22. Vessel density 500  m 100  m Harrison RV et al. Cerebral cortex. 2002

23. Arteriole 100  m

24. Even smaller 50  m

25. Arterial side  Capillaries are randomly orientated  Oxygen exchange in capillaries  Arterioles perform local CBF control Artery Blood oxygen saturation, 98-100% Arterioles 25  m 15% of CBV Capillaries 8  m 40% CBV

26. Venous side  Venules  are (approx) randomly orientated  have the same blood volume as capillaries  have twice the deoxyHb concentration of capillaries  are more (para)magnetic than capillaries and arteries Vein Blood oxygen saturation (resting), 60% Venules 25-50  m 40% of CBV Capillaries

27. Activation Rest Active: 50% increase in CBF, 20% increase in CMRO2 O2 Sat100%80%60% O2 Sat100%86%72%

28. Decrease in deoxy-Hb concentration

29. BOLD FMRI Basal (resting) state capillary bed venules arterioles = HbO 2 = Hbr FLOW CBV Field gradients capillary bed venules arterioles = HbO 2 = Hbr MRI signal electrical activity hemodynamic response xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx - normal flow - basal level [Hbr] - basal CBV - normal MRI signal

30. BOLD FMRI capillary bed venules arterioles = HbO 2 = Hbr FLOW CBV Activated state capillary bed venules arterioles = HbO 2 = Hbr FLOW CBV MRI signal electrical activity hemodynamic response xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx - increased flow - decreased [Hbr] - increased CBV - increased MRI signal

31. Dissecting BOLD S BOLD =f(CBV,CBF,CMRO 2 ) Buxton et al. Neuroimage 2004 Purer measures of neuronal activity?

32. FMRI Modelling: The Haemodynamic Response  The stimulus is convolved with an assumed or modeled impulse response function, the haemodynamic response function (HRF), to give the assumed BOLD response

33. HRF Stimulus (Single Event) Haemodynamic Response Function (HRF) 0 10 20 30 seconds The haemodynamic response to a stimulus is blurred and delayed Time

34. Predicted Response  The process can be modeled by convolving the activity curve with the HRF HRF Predicted neural activity Predicted response   =

35. GLM  Standard GLM Analysis:  Correlate model at each voxel separately  Measure residual noise variance  T-statistic = model fit / noise amplitude  Threshold T-stats and display map  Signals of no interest (e.g. artifacts) can affect both activation strength and residual noise variance  Use pre-processing to reduce/eliminate some of these effects

36. A typical scan at 3T…  Gradient echo EPI  TR 2000 – 4000 ms  TE 30 ms  Flip angle 60 – 90 degrees  64 x 64  4 – 8 mm thick slices  20 – 40 slices

37. Why EPI?  A typical T2-weighted imaging series requires that TR be two to three times longer than the intrinsic tissue magnetization parameter, T1.  The T1 of biological samples is typically on the order of a second; TR must therefore be 3 seconds or more.  A typical MR image is formed from 128 repeated samples, so that the imaging time for our canonical T2 weighted scan is about 384 seconds, or more than 6.5 minutes.  By comparison, the EPI approach collects all of the image data, for an image of the same resolution, in 40 to 150 milliseconds  A nearly 10,000-fold speed gain!

38. However…  Bandwidth and Artifacts  Chemical Shift  Shape Distortion  Ghosting  Resolution

39. Purer physiological measures  Perfusion and perfusion change  CMRO2 change  Cerebral blood volume  Oxygen extraction fraction

40. neuronal activity - excitatory - inhibitory - soma action potential metabolic response -  glucose consumption -  oxygen consumption hemodynamic response -  blood flow -  blood volume -  blood oxygenation autoradiography FDG PETNIRSH 2 15 O PET FMRI optical imaging electrophysiology EEG MEG Neurovascular coupling Correlates of brain activity

41. Oxidative metabolism attenuates BOLD signal  Hoge R et al CBF BOLD

42. CMRO2 measurement R2* (BOLD) = k CBV  [dHb]  k = field dependent constant CBV = cerebral blood volume fraction [dHb] = concentration of dHb in blood  = theoretical CBV dependence (  =1)  = theoretical [dHb] dependence –   1.5 (1.5T) [Boxerman et al, 1995] –   1 (>3T) [Ogawa et al, 1993] Measured BOLD

43. [dHb] = CMRO 2 / CBF Substitutions: (Fick’s principle) CBV (Grubb et al., 1974)  = 0.38 (steady state value) CBV 0 CBF 0 CBF   =  CMRO 2 = CBF. OEF. C a ASL measured CMRO 2 measurement

44.  R2* (BOLD) R2* 0(BOLD) CMRO 2(0) CMRO 2    CBF 0 CBF     1 - ] = Calibrate R2* 0 using a hypercapnia challenge CMRO 2 measurement A flow increase without increase in CMRO 2

45. Calibrated BOLD for measuring CMRO 2 Hoge R et al CMRO2-CBF coupling: slope ~2 Calibrated BOLD

46. What else?  ASL  VASO  MRS  DWI  ???

47. Neural Response Stimulus CBV CBF CMRO 2 BOLD Neurovascular Coupling

48. Arterial Spin Labeling Williams et al. PNAS 1992 Relies on endogenous contrast agent Magnetically label water at the neck (below imaging plane) Labeled blood moves downstream and mixes with stationary tissue water movement of blood water: measure of CBF CBF (ml g -1 min -1 ) pCO 2 (mmHg)

49. ASL: Tagging Strategies Universal concept: “tag” and “control” tag image Wait for tag blood water spins to arrive tag delay: 500 - 2000 msec time

50. ASL: Tagging Strategies Universal concept: “tag” and “control” Wait the same amount of time but no tagging no tagimage

51. ASL: Tagging Strategies Universal concept: “tag” and “control” no tagimage Control image is important same tag location off-resonance global tag Wait the same amount of time but no tagging

52. A bit about baselines hemodynamic response

53. Physiological baseline  Baseline CBF ,  But  CBF  CMRO 2 unchanged (probably) (Brown et al JCBFM 2003)  BOLD response  (probably) Cohen et al JCBFM 2002

54. Implications  Factors altering baseline state  Disease  Sedation  Anxiety  Vasoactive medications  Global and local   CBF (ASL) may be more robust?

55. Noise sources  What is noise in a BOLD experiment?  Unmodelled variation in the time-series  Intrinsic MRI noise  Independent of field strength, TE  Thermal noise from subject and RF coil  Physiological noise  Increases with field strength, depends on TE  Cardiac pulsations  Respiratory motion and B0 shift  Vasomotion, 0.1Hz  Blood gas fluctuations  “Resting state” networks  Also  Scanner drift (heating up)

56. At 3Tesla Physiological noise > scanner + thermal noise Physiological noise GM > Physiological noise WM

57. Spatial distribution of noise  Motion at intensity boundaries  volunteer  Respiratory B0 shift  Physiological noise in blood vessels and grey matter

58. Noise structure  1/f dependence  BOLD is bad for detecting long time-scale activation frequency BOLD noise

59. Noise or signal?  Noise is unmodelled signal  Spatially structured  Temporally structured  “Physiological” signal  Vascular properties  “Neuronal” signal  Resting state networks  Resting fluctuations  Stimulus induced deactivation Separation: all haemodynamic

60. Physiological noise  Motion  McFLIRT correction  Cardiac  Pulsations (aliased)  Respiratory  Motion  B 0 shift RETROICOR correction

61. Physiological signal  Low frequency haemodynamic oscillations  Information about vascular properties  CO 2 reactivity  Autoregulation Is it a problem? Can we use it?

62. BOLD response to CO 2 CO 2 is a potent vasodilator 50 100 CBF ml/100g/min Normal Pa CO2 80604020mmHg CBV ml/100g 4.0 5.0 3.0 Hypercapnia: CBF, CBV   [deoxyHb]   T2*   S BOLD  Previous investigations use sustained hyper/hypocapnia challenges to investigate regional sensitivity (1.5T) e.g. Posse et al. 1997, 2001, Rostrup et al. 2000

63. Spontaneous CO 2 fluctuations Resting P ET CO 2 P ET CO 2 power spectrum End-tidal CO 2 (P ET CO 2 ) is a good measure of arterial CO 2 Fluctuations 0 - 0.05 Hz ( Van den Aardweg & Karemaker, 2002) Overlaps with stimulus frequencies Can correlate with stimulation Wise et al Neuroimage 2004

64. BOLD-CO 2 (resting) correlation r=0.5, Z=5.5

65. What about 7T?

66. FMRI at 7T

67. Bigger is better  High field = high SNR and high BOLD CNR  At higher magnetic fields, the short T2 of blood means that its signal is attenuated relative to that from tissue at the TE’s used for fMRI.  Hence it is expected that higher spatial specificity can be obtained in BOLD data acquired at high field as the intra-vascular signal contribution from draining veins is reduced

68. Pros and Cons  The increase in physiological noise can reduce the gains in BOLD CNR at ultra-high field  Higher BOLD signal means more signal drop-out  However, an increased BOLD CNR allows fMRI to be performed at higher spatial resolution, which has the potential advantage of reduced signal drop-out due to magnetic field inhomogeneities and longer baseline T 2  At high resolution the effect of intrinsic noise can dominate that of physiological noise

69. Extra Reading: Buxton et al. Modeling the hemodynamic response to brain activation. NeuroImage 23 (2004) S220–S233 Raichle & Mintun. BrainWork and Brain Imaging. Annu. Rev. Neurosci. 2006. 29:449–76 http://www.fmrib.ox.ac.uk/Members/bulte/