Skip to Main Content
We present a novel multivariate machine learning approach to the identification of voxel clusters containing brain state discriminating information, serving as a potentially more sensitive alternative to univariate activation detection. The proposed method consists of an evolutionary algorithm that, in conjunction with a classifier, extracts voxel clusters with a classification score above a pre-defined, above-chance threshold. The results can be displayed as two- or three-dimensional voxel discrimination relevance maps (VDRMs), indicating where and to what degree brain state classification is possible. When applied to a finger-tapping dataset numerous voxel clusters with impressive classification rates were identified, at best scoring an area under the receiver operating characteristic curve (ROC)-curve (AUC) of 1 within as well as between subjects. The location of high-scoring regions correlated well with functionally relevant areas as defined by the general linear model (GLM). Combining clusters for maximal classification scores as a feature selection approach outperformed the GLM T-map voxel ranking method (e.g., group level AUC of 0.908 compared to 0.785 for one cluster/200 voxels). Moreover, on data from a tactile study we show that the proposed algorithm can produce significant brain state discrimination scores where both the GLM and ROI-based classification fail to detect significantly activated voxels. Finally, we demonstrate that the algorithm can be successfully applied to data with more than two conditions and hence produce multiclass voxel relevance maps. The proposed evolutionary classification scheme has thus proven excellent in identifying voxel clusters that contain information about given brain states, which can be utilized not only for maximal single-volume fMRI classification, but also for multivariate, multiclass, highly sensitive functional brain mapping.