Classification of fMRI Signals Using Brain Activation-Based Decision Tree

Spatio–temporal activation of different brain regions characterizes a task performed by an individual. The consistency of the activation and the dynamics of the same between the brain regions reveal the functional relationship among the regions. In this letter, we propose a novel method to classify functional magnetic resonance imaging (fMRI) data captured while performing certain cognitive tasks. In the present analysis, four tasks are used, namely, visual word and objects, visual working memory, Simon task, and tone counting. A generalized temporal consistency in activation level (GTCAL) of the brain regions is proposed as a feature to construct a decision tree for the classification. Hammersmith atlas is used for parcellation of the brain regions. The decision tree-based approach enables to overcome the ambiguity arising due to partial similarity in the activation/deactivation of pairwise brain regions for different tasks. The proposed approach is found to outperform the existing methods of classification, namely, cross correlation, Euclidean distance, city block distance, Wasserstein distance, and temporal consistency in activation level. Such a classification of fMRI data for different functional tasks is useful to characterize the effect of aging in healthy population and to study various neurological disease conditions.