Functional magnetic resonance imaging (fMRI) time series generally demonstrate serial dependence. This endogenous auto-correlation typically exhibits long-range dependence described by a 1/f-like power law. We present a novel wavelet-based methodology for characterising the noise structure in short-medium length (shortish) fMRI time series. Mono-fractality is assessed in terms of the Hurst exponent and the noise variance. We then investigate potential local stationarity of the Hurst exponent in MM data and present a uniformly most powerful test for its time constancy. A novel bootstrap approach is presented as an alternative to the normal assumption and its advantages are discussed. From several datasets investigated, we specifically show that the 1/f model is particularly suited to describe color in MM nose. We also demonstrate that even if most of the brain voxels are mono-fractal, there are many locations in the brain where time constancy of the Hurst exponent is violated, ie, the noise structure is poly-fractal.
[1]
M. D’Esposito,et al.
Empirical Analyses of BOLD fMRI Statistics
,
1997,
NeuroImage.
[2]
E. Lehmann,et al.
Testing Statistical Hypothesis.
,
1960
.
[3]
A. Oppenheim,et al.
Signal processing with fractals: a wavelet-based approach
,
1996
.
[4]
P. Hall.
The Bootstrap and Edgeworth Expansion
,
1992
.
[5]
T A Carpenter,et al.
Colored noise and computational inference in neurophysiological (fMRI) time series analysis: Resampling methods in time and wavelet domains
,
2001,
Human brain mapping.
[6]
M. D’Esposito,et al.
Empirical analyses of BOLD fMRI statistics. I. Spatially unsmoothed data collected under null-hypothesis conditions.
,
1997,
NeuroImage.