Wavelet bootstrapping estimation of fMRI data
Dr. Maria Gavrilescu
Email: maria(at)pcomm.hfi.unimelb.edu.au
The effective connectivity analysis of fMRI data allows us to get an insight of how different brain regions work together in order to perform a given task. One requirement in many estimation algorithms is the multivariate normality of the data. From our previous work we observed that this condition is often violated for fMRI data. Under these circumstances, re-sampling techniques such as bootstrapping may provide a better alternative for analysis. However, bootstrapping the fMRI data is a challenging problem since the fMRI time series exhibit strong autocorrelation properties. Bootstrapping applied directly to the fMRI time series will disturb the autocorrelation structure and will therefore render samples that are not equivalent. By performing a wavelet decomposition of the data and applying bootstrapping in the wavelets domain the autocorrelation structure of the data can be preserved.
The aim of this project is to implement and investigate the use of wavelet bootstrapping for effective connectivity estimation in fMRI data. Different approaches on data modelling (based on autoregressive models for examples) will also be explored.