PhD Thesis Final Defense to be held on September 10, 2019, at 17:00
Photo credit: Antonis Savva
The examination is open to anyone who wishes to attend (Multimedia Room 1, Central Library of NTUA).
Thesis Title: Methodologies for Assessing Dynamic Functional Connectivity in Functional Magnetic Resonance Imaging
Abstract: Recent studies related to assessing functional connectivity in resting-state functional Magnetic Resonance Imaging have revealed that the resulting connectivity patterns exhibit considerable fluctuations (dynamic functional connectivity), contrary to earlier assumptions of stationarity. A widely applied method for quantifying dynamic functional connectivity is the sliding window method. According to this method, the data are divided into segments with the same length (window size) and a correlation metric is employed to assess connectivity within these segments, whereby the window size is often empirically chosen. Usually, the data points inside each window are given equal weights; however, it has been argued that this approach could result in abrupt changes in windowed metric series if an outlier is considered. For minimizing sensitivity to outliers a variety of window functions were employed in the corresponding literature.
In the present thesis, the assessment of dynamic functional connectivity from resting-state functional Magnetic Resonance Imaging data, is rigorously investigated using the sliding window method. Firstly, to fully examine the effects of weighting in the sliding window method, the present thesis accomplishes an exhaustive investigation using ten different window functions. Secondly, a detailed comparison is performed between different correlation metrics, across a wide range of window sizes, aiming to systematically define an optimum parameter selection to assess dynamic functional connectivity. Statistical inference is achieved by employing a hypothesis testing framework, based on surrogate data, for constructing the null hypothesis of dynamic functional connectivity absence. Based on this analysis, we sought to identify the sensitivity of each metric and window function with respect to window size, for identifying dynamic functional connectivity.
Subsequently, in the present thesis, another method for estimating dynamic functional connectivity is considered, whereby the time-series are projected to the time-frequency domain. This technique is based on the wavelet transform for obtaining a coherence measure of the examined signals in both time and frequency, as well as their phase coupling values. Next, characteristic curves are estimated, indicating the most prominent wavelengths in dynamic functional connectivity. Using both visual illustrations and a hypothesis testing framework, based on surrogate data, a broader interpretation of dynamic functional connectivity is achieved, which only requires the selection of the wavelet basis function for the transform, as opposed to the sliding window method that requires the definition of the functional connectivity metric, window size, window shifting and window function. Consequently, the conclusions depend on fewer methodological parameters and can provide a better understanding of functional connectivity patterns of the resting brain.
Supervisor: George Matsopoulos, Professor
PhD student: Antonis Savva