Comparing univalent and bivalent brain functional connectivity measures using machine learning

Functional connectivity is the stochastic association or the dependency of two or more distinct brain regions. It is primarily used for finding patterns that are validated through statistical methods, in the context of brain connectivity. Quantification of functional connectivity is usually performed using Pearson's correlation coefficient (PCC). Many Functional magnetic resonance imaging (fMRI) studies have used PCC to quantify functional connectivity in a bivalent sense. However, the interpretation of negative fMRI responses or deactivation has proved challenging. Therefore, few have employed the absolute value of PCC (univalent) to model functional connectivity. This paper compares the two measures and assesses their performance and suitability for fMRI connectivity modeling. Connectivity analysis and classification of autistic individuals from control population is performed using these two measures. Machine learning classification is employed to quantify the predictive abilities of univalent and bivalent functional connectivity measures. This paper experimentally finds the usage of bivalent measure to be producing better classification accuracy by around 2%, which means it is more suitable for fMRI functional connectivity analysis.