A Data-Driven fMRI Analysis Method Using Connected Components and K-Means Algorithm

Functional magnetic resonance imaging (fMRI) is an increasingly used method for studying human brain activity changes in vivo. For analysing data which are lack of expected response and brain regions of interest, we have previously proposed a completely data-driven method in subject level, which sought for spatially connected voxels with their temporal maxima occurring within a detected time window, in the whole brain. As subject number grows, the interpretation of various numbers of components obtained, depending on the temporal peaks and their spatial associations, as a whole became extremely difficult. Here, we propose a group-level analysis method, which utilises the connected components from all subjects and the K-means algorithm, and returns the period during which there is a significant response and an activation map. For validation, four data sets acquired in a single-event visual experiment were used and the associated time window and brain regions to the experimental paradigm were detected.