Creating Group-Level Functionally-Defined Atlases for Diagnostic Classification

In this paper we introduce a method to produce a subdivision of an anatomical atlas by taking into account the similarity of resting state functional MRI time series within anatomically-defined regions of interest. This method differs from others in that the resulting atlases are comparable across subjects, making group analyses possible. Finally, we show that the functional connectivity matrices obtained with this method can be used in a diagnostic classification task and that they enhance a classifier's ability to extract relevant information from the data, leading to more interpretable prediction models in the process.