Detecting low-frequency functional connectivity in fMRI using unsupervised clustering algorithms

Abstract

Recent research in functional magnetic resonance imaging (fMRI) revealed slowly varying temporally correlated fluctuations between functionally related areas. These low-frequency oscillations of less than 0.08 Hz appear to be a property of symmetric cortices, and they are known to be present in the motor cortex among others. These low-frequency data are difficult to detect and quantify in fMRI. Traditionally, user-based regions of interests (ROI) or "seed clusters" have been the primary analysis method. We propose in this paper to employ unsupervised clustering algorithms employing arbitrary distance measures to detect the resting state of functional connectivity. There are two main benefits using unsupervised algorithms instead of traditional techniques: (1) the scan time is reduced by finding directly the activation data set, and (2) the whole data set is considered and not a relative correlation map. The achieved results are evaluated for different distance metrics. The Euclidian metric implemented by the standard unsupervised clustering approaches is compared with a more general topographic mapping of proximities based on the correlation and the prediction error between time courses. Thus, we are able to detect functional connectivity based on model-free analysis methods implementing arbitrary distance metrics.