Classification and Visualization of Multiclass fMRI Data Using Supervised Self-Organizing Maps

So far, most fMRI studies that analyzed voxel activity patterns of more than two conditions transformed the multiclass problem into a series of binary problems. Furthermore, visualizations of the topology of underlying representations are usually not presented. Here, we explore the feasibility of different types of supervised self-organizing maps (SSOM) to decode and visualize voxel patterns of fMRI datasets consisting of multiple conditions. Our results suggest that - compared to commonly applied classification approaches - SSOMs are well suited when activity patterns consist of a small number of features (e.g. as in searchlight- or region of interest-based approaches). In addition, we demonstrate the utility of using SOM grids for intuitive and exploratory visualization of topological relations among classes of fMRI activity patterns.