Nonwhite noise in fMRI: Does modeling have an impact

1
Non-white noise in fMRIDoes modeling have an
impact

• By Lund, Madsen, Sldaros, Luo and Nichols in
  NeuroImage (2006)
• Jul. 27, 2006

2
GOAL

• 1. To remove temporal autocorrelation remaining
  even after whitening step (rigid body
  transformation)
• ? Nuisance Variable Regression (NVR)
• 2. To demonstrate the modeling effect by some
  diagnostic tools with some data sets, e.g.,
  simulated data, phantom data.

3
Noises and the effect on true signal

• Possible noises Low frequency drift due to
  hardware imperfection, oscillatory noise due to
  respiration and cardiac pulsation, residual
  movement artefacts
• These noising factors induce temporal
  autocorrelation, invalidate statistical analysis
• since statistical analysis assume the residuals
  are independent and identically distributed
  normal

4
Background knowledge in noise

• Low frequency oscillation is nature which is
  found in cadaver and phantom.
• Typically, in brain acquisition, the
  physiological noise components are heavily
  aliased and non-stationary ? No commonly accepted
  standard noise model.

5
Background knowledge in noise (Contd)

• A real signal S is sampled at frequency
• Then, the frequency (detected from the
  sampled signal), which is
• is aliased with the frequency
• - Nyquist theorem

6

• Nonwhite noise sources
• 1. Low frequency drift due to hardware
  instability
• Commonly used model is to include a basis set of
  slowly varying functions (in fmristat by Worsley
  and Friston, 1995)
• Discrete cosine set (in SPM2)

7
(No Transcript)
8

• 2. Residual Movement effect which can be present
  in the data even after rigid body transformation
  (Friston et al. 1996)
• m_i(t) is the 6 rigid body movement parameter of
  the volume at time t_n

9

• 3. Modeling aliased physiological noise
• RETROICOR method, (Glover et al. (2000)) models
  the physiological noise as a basis set of sine
  and cosine.

10

• Where are coefficeints of sines and
  cosines describing the five harmonics of the
  cardiac noise and
• are the coefficients of sines and cosines
  of describing 3 harmonica of the respiratory noise

11

• RETROICOR method

12
(No Transcript)
13
Nuisance Variable Regreesion

• Nuisance Variable Regression (NVR) is
• obtained by combining three components of

14
(No Transcript)
15
(No Transcript)
16
Diagnostics

• Testing assumptions in general linear model
• Testing on whiteness and normality
• By using Statistical Parametric Mapping
  diagnostics (SPMd)
• Test statistics
• BLUS residuals, Durbin-Watson (whiteness)
• Shapiro and Wilk (normality)

17
Next presentation!

• Investigation into resting state connectivity
  using Independent component analysis
• By Beckmann, et al. (2005)
• It allows the scientific investigation in order
  to characterize the identification of low
  frequency resting state patterns