Increasing robustness of pairwise methods for effective connectivity in Magnetic Resonance Imaging by using fractional moment series of BOLD signal distributions

Estimating causal interactions in the brain from functional magnetic resonance imaging (fMRI) data remains a formidable task. Multiple studies have demonstrated that essentially all current analytical approaches to determine effective connectivity perform poorly even when applied to synthetic fMRI datasets. Recent advances in this field include methods for pairwise inference, which involve creating a sparse connectome in the first step of the inference, and then using a classifier in order to determine the directionality of every connection in the second step. In this work, we propose an advance to the second step of this inference procedure, by building a classifier based on all moments of the BOLD distribution combined into cumulants. The classifier is trained on a dataset generated under under the Dynamic Causal Modeling (DCM) generative model (Friston et al, 2003). First, we evaluated the performance of this classifier on synthetic benchmark datasets. Secondly, we simulated the DCM generative model with natural confounds present in the fMRI datasets such as varying signal magnitudes and noise variances between the upstream and downstream regions. Our approach outperforms other methods for effective connectivity research when applied to the benchmark datasets, but crucially, it is also more resilient to known confounding effects such as differential noise level across different areas of the connectome. This suggests that the classifier proposed in this work can be recommended for further validation on the human fMRI datasets.