Thesis defense Natalia Bielczyk (Donders series 432)

18 June 2020

Promotors: prof. dr. C. Beckmann, prof. dr. J. Buitelaar
Co-promotor: dr. J. Glennon

Signal detection and causal inference in functional Magnetic Resonance Imaging

How to make a map of causal connections between brain areas on the basis of functional Magnetic Resonance Imaging (fMRI) datasets? The thesis summarizes the state of the art methodology for signal detection and causal inference in fMRI, and introduces Several original contributions to the field. One common feature of the proposed methods is that they are based on analyzing the properties of the BOLD fMRI time series compressed into distributions of BOLD fMRI values. This compromise is a consequence of the fact that BOLD fMRI time series has a poor temporal characteristics due to the slow hemodynamics.

Firstly, “Momentum”, a method for finding a signal in fMRI datasets by detecting nongaussianity in the distribution of BOLD fMRI values, is introduced. This method can serve for charactering brain activity at rest, and help in improving on efficiency of the functional methods for brain parcellation. As a test for nongaussianity, Momentum can also be used in multiple other branches of research beyond neuroimaging.

Secondly, a method for detecting significant connections in functional connectomes using the concepts of mixture modeling and the False Discovery Rate, is introduced. As demonstrated in the thesis, this method can be an effective alternative to permutation testing, especially useful in small datasets.

Lastly, a method for determining the structure of the causal connectome based on fMRI data using fractional moments of the distribution combined into cumulants, is introduced. This method can help in creating more reliable causal connectomes based on fMRI data as it is more resilient to confounders naturally occurring in the brain networks than the state-of-the-art methods for causal inference.