Analyzing fMRI data with graph-based visualizations of self-organizing maps

Self-organizing mapping (SOM) is a topology-preserving unsupervised manifold learning technique that maps high-dimensional data into a low-dimensional (often a 2-D) space. SOM has been successfully used as a data-driven approach for model-free functional magnetic resonance imaging (fMRI) data analysis. However, effective clustering or interpretation of the prototypes (weight vectors) in the map is necessary to delineate fine cluster structures and features of interest in the data. In this work, we used graph-based visualization techniques to capture neighborhood relations among the SOM prototypes based upon (i) distribution of data across the receptive fields of the prototypes and (ii) temporal similarities (correlations) in the prototypes. These help in advanced visualization of cluster boundaries in fMRI data enabling the separation of regions with small onset differences (delays) in the blood oxygenation level-dependent (BOLD) responses in visual cortex and possibly other regions of brain.