Daniel Bulte Centre for Functional Magnetic Resonance Imaging of the Brain University of Oxford
Overview Signal Sources BOLD Contrast Mechanism of MR signal change FMRI Modelling Scan design details Factors affecting BOLD Changing physiological baseline Noise Other modalities ASL & more
Activation
Deoxy-Haemoglobin paramagnetic different to tissue Δχ=0.08ppm
Oxy-Haemoglobin diamagnetic same as tissue
Field homogeneity & oxygenation state Red blood cell 6 µm diameter, 1 2 µm thick Susceptibility An object with differing magnetic properties distorts the field
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
Magnetic field in a vessel B 0 θ a r φ Inside Cylinder ω(θ)=δω [3cos 2 (θ)-1]/3 Δ ω ' = 2π(1-Y) Δχ '
Magnetic field around a vessel B 0 θ a Outside Cylinder Δω(r,θ,φ)= Δω sin 2 (θ)(a/r) 2 cos(2φ) r φ Inside Cylinder ω(θ)=δω [3cos 2 (θ)-1]/3 Δ ω ' = 2π(1-Y) Δχ '
Vessel orientation Field inside and outside depends on angle θ with respect to B0 Bandettini and Wong. Int. J. Imaging Systems and Technology. 6:133 (1995)
Blood oxygenation Field inside and outside depends on Y, oxygenation Bandettini and Wong. Int. J. Imaging Systems and Technology. 6:133 (1995)
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 ΔR2* = 4.3 γδχ(1-y) B 0 CBV (venules, larger vessels) ΔR2* = 0.04 {γδχ(1-y)} 2 B 0 2 CBV (smaller capillaries)
BOLD signal Blood Oxygen Level Dependent signal T2* change from the haemodynamic perturbation associated with neural activation
From neural activity to BOLD signal Neural activity Signalling Vascular response BOLD signal Synaptic signalling Vascular tone (reactivity) Autoregulation Blood flow, oxygenation and volume arteriole B 0 field glia Metabolic signalling venule
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
Haemodynamic changes underlying BOLD CBF CBV BOLD response, % 3 2 1 0 initial dip positive BOLD response overshoot post stimulus undershoot CMRO 2 stimulus time stimulus
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?
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)
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 ΔR2* (1-Y) β CBV Gradient echo Y=O 2 saturation β~1.5 S = S max. e -TE/R2*
Echo time and BOLD sensitivity BOLD contrast-to-noise optimised when TE~T2* T2* shorter at higher field Relative CNR TE (ms) TE optimization similar to T2 structural optimizations, just between different states rather than different tissues
Vessel density 500 µm 100 µm Harrison RV et al. Cerebral cortex. 2002
Arteriole 100 µm
Even smaller 50 µm
Arterial side Capillaries 8 µm 40% CBV Arterioles 25 µm 15% of CBV Capillaries are randomly orientated Oxygen exchange in capillaries Arterioles perform local CBF control Artery Blood oxygen saturation, 98-100%
Venous side Capillaries Venules 25-50 µm 40% of CBV Vein Blood oxygen saturation (resting), 60% 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
Rest Activation O2 Sat 100% 80% 60% Active: 50% increase in CBF, 20% increase in CMRO2 O2 Sat 100% 86% 72%
BOLD FMRI Basal (resting) state MRI signal arterioles CBV capillary bed venules - normal flow - basal level [Hbr] - basal CBV - normal MRI signal FLOW electrical activity Field gradients = HbO2 = Hbr hemodynamic response x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x
BOLD FMRI Activated state MRI signal arterioles CBV capillary bed venules increased flow decreased [Hbr] increased CBV increased MRI signal FLOW = HbO2 = Hbr electrical activity hemodynamic response x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x
Dissecting BOLD S BOLD =f(cbv,cbf,cmro 2 ) Purer measures of neuronal activity? Buxton et al. Neuroimage 2004
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
HRF Stimulus (Single Event) Haemodynamic Response Function (HRF) 0 10 20 30 seconds Time The haemodynamic response to a stimulus is blurred and delayed
Predicted Response The process can be modeled by convolving the activity curve with the HRF HRF Predicted neural activity = Predicted response
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
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
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!
Bandwidth and Artifacts Chemical Shift Shape Distortion Ghosting Resolution However
hemodynamic response A bit about baselines
Physiological baseline Baseline CBF, But ΔCBF ΔCMRO 2 unchanged (probably) (Brown et al JCBFM 2003) BOLD response (probably) Cohen et al JCBFM 2002
Implications Factors altering baseline state Disease Sedation Anxiety Vasoactive medications Global and local ΔCBF (ASL) may be more robust?
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)
At 3Tesla Physiological noise > scanner + thermal noise Physiological noise GM > Physiological noise WM
Spatial distribution of noise Motion at intensity boundaries volunteer Respiratory B0 shift Physiological noise in blood vessels and grey matter
Noise Structure BOLD noise frequency 1/f dependence BOLD is bad for detecting long time scale activation
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
Physiological noise Motion McFLIRT correction Cardiac Pulsations (aliased) Respiratory Motion B 0 shift RETROICOR correction
Physiological signal Low frequency haemodynamic oscillations Information about vascular properties CO 2 reactivity Autoregulation Is it a problem? Can we use it?
Purer physiological measures Perfusion and perfusion change CMRO2 change Cerebral blood volume Oxygen extraction fraction
Correlates of brain activity electrophysiology EEG MEG metabolic response - glucose consumption - oxygen consumption FDG PET autoradiography neuronal activity - excitatory - inhibitory - soma action potential hemodynamic response - blood flow - blood volume - blood oxygenation H 2 15 O PET NIRS optical imaging Neurovascular coupling FMRI
Cerebral Blood Flow Neurovascular Coupling
Arterial Spin Labelling CBF (ml g -1 min -1 ) pco2 (mmhg) 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
ASL: Tagging Strategies Universal concept: tag and control tag image Wait for tag blood water spins to arrive tag delay: 500-2000 msec
ASL: Tagging Strategies Universal concept: tag and control no tag image Wait the same amount of time but no tagging
ASL: Tagging Strategies Universal concept: tag and control Control image is important same tag location off-resonance no tag image global tag Wait the same amount of time but no tagging
What else? ASL VASO MRS DWI???
http://www.fmrib.ox.ac.uk/members/bulte/ 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