Quantifying nonlinear coupling in cerebral haemodynamics using information transfer
Journal
Journal of Cerebral Blood Flow and Metabolism
Journal Volume
27
Journal Issue
SUPPL. 1
Pages
BP04-05W
Date Issued
2007
Author(s)
Abstract
BACKGROUND: The information theoretical concept of transfer entropy [Schreiber, T. Phys Rev Lett. 2000] was recently used [Katura et al. NeuroImage. 2006.] to assess the influence of systemic variability on spontaneous low frequency oscillations measured in the human brain using diffuse optical tomography. Transfer entropy is a probabilistic technique of assessing general coupling between time series. It has considerable advantages over traditional techniques (such as correlation) as it makes no assumptions about the linearity of the underlying relationship. It was recently shown [Achard, S. Journal of Neuroscience. 2006] that correlation of coefficients of the maximal overlap wavelet packet transform of functional magnetic resonance imaging (fMRI) time series in humans can be used to characterise functional connectivity between regions of the brain by assessing similarity between time series of low frequency (<0.1Hz) oscillations. Importantly, the wavelet packet transform provides a means of separating the time series into components at different scales (frequency bands). When this decomposition has been performed, correlation provides a measure of linear similarity between scales. Replacing this correlation step in the algorithm with a similarity measure based upon transfer entropy relaxes the assumption of linearity and can be used to estimate a directed measure of information transfer between inter scale time series. METHODS: We have developed an estimator for transfer entropy that uses a recently developed unsupervised learning algorithm [Figueiredo et al. IEEE Trans. on Pattern Analysis and Machine Intelligence 2002.] for estimation of the probability density functions of the time series embedding from relatively short duration records. This estimator is applied to quantification of the interrelationship between wavelet transform coefficients of low frequency oscillations measured using near infrared spectroscopy (NIRS) and those observed systemically (for example mean arterial blood pressure (ABP)). RESULTS AND CONCLUSIONS: Use of a Gaussian mixture model to converge on the underlying probability density functions provides a practical estimator for transfer entropy between time series that may be measured using near infrared spectroscopy. The maximal overlap wavelet packet transform is an efficient algorithm for decomposing a time series into coefficients that have highly localised frequency content. Applying the technique to quantification of coupling between NIRS and systemically measured variability provides a measure of the extent to which low frequency behaviour measured using NIRS can be attributed to systemic variability. It is expected that this technique will also be useful to investigate functional connectivity between regions of the brain by assessing scale dependent similarity of low frequency oscillations measured using FMRI without making the assumption of linearity of the underlying coupling relationship.
Subjects
algorithm
arterial pressure
brain function
brain region
conference paper
entropy
functional magnetic resonance imaging
hemodynamics
mathematical model
near infrared spectroscopy
priority journal
Type
conference paper