fMRI Signal, Noise and Experimental Design

1
fMRI Signal, Noise and Experimental Design

• Torben Ellegaard Lund

2
The BOLD signal

• Increased neural activity leads to
• Increased regional oxygen consumption (rCMRO2)
• Increased regional cerebral blood flow (rCBF)
• Increased regional cerebral blood volume (rCBV)
• A decrease in the deoxyhaemoglobin concentration
  leads to a signal increase in T2 and T2 weighted
  images

3
The BOLD signal

• The three components has different time-constants
  and effects
• The rCMRO2 increase leads to a fast but small
  signal decrease known as The initial dip (1s)
• The rCBF increase leads to the peak which usually
  dominates the haemodynamic response function (6s)
• The rCBV increase leads to a signal
  undershoot.(12s)

4
The BOLD signal

• What are the optimal imaging parameters?
• TE, Bandwidth, sequence type
• Spatiotemporal resolution/coverage
• Slice orientation
• Etc.

5
The BOLD signal

• T2 or T2 ?

6
The BOLD signal

• Which TE?
• S(t)I0e-t/T2

7
The BOLD signal

• A time course

Langkilde Rostrup
8
Noise

• There are several other things than the BOLD
  signal which vary during a fMRI time course
• Drift in shim and gradient currents (1/f noise)
• Cardiac pulsation
• Respiration
• Swallowing
• Head motion
• Eye movement
• Stimulus locked motion

9
1/f Noise

• Suggested origins
• Drift in shim and gradient currents
• Motion
• Spontaneous neural activity

10
1/f Noise

• Solutions

• Fitting T2

• Blocking events
• together

11
Stimulus locked motion
Birn et al. HBM 1999

• Examples
• Swallowing
• Speaking
• Jaw clenching
• Tongue movement
• Eye movements
• Head movement

12
Cardiac Induced Noise

• When critically sampled e.g. TR143ms

Weisskoff et al. 1993
13
Aliasing

• When?
• Nyquist criterion
• 2f ? fs
• Example heart rates
• 61, 74 and 91 bpm
• Sampled at 0.5Hz
• i.e. TR2s

14
Jitter

• The heart rate is not constant (fs5.4Hz)

?f0.1Hz6bpm
15
Cardiac Induced Noise

• When not critically sampled
• e.g. TR3000ms

Hu et al. 1993 Dagli et al. 1999
16
Cardiac Induced Noise

• GLM regressors
• Paradigm and 5 vessel time series (per session)
• F-test
• Null-hypothesis No effect of vessel time series
• plt0.05 corrected

Lund et al. ISMRM 2002
17
Cardiac Induced Noise

• GLM regressors
• Paradigm and 5 vessel time series (per session)
• F-test
• Null-hypothesis No effect
• of vessel time series
• plt0.05 corrected

Lund et al. ISMRM 2002
18
Cardiac Induced Noise

• Significant effect of nuisance covariates in
• Areas surrounding branches of the medial-,
  posterior and anterior cerebral arteries
• Ventricles
• The dural venous sinuses

Lund et al. ISMRM 2002
19
Cardiac Induced Noise
Example Flickering checkerboard

• Over sampled TR200ms

• Under sampled TR1.6s

• Under sampled TR1.6s
• -but cardiac noise is now modelled

Lund et al. ISMRM 2001
20
Links

• SPM http//www.fil.ion.ucl.ac.uk/spm/
• FSL http//www.fmrib.ox.ac.uk/fsl/index.html
• SPM extensions http//www.fil.ion.ucl.ac.uk/spm/e
  xt/
• DRCMR http//www.drcmr.dk/