Utilization of Spatial Coherence in Functional Neuroimage-Based Classification

Functional magnetic resonance imaging provides a non-invasive mechanism for monitoring brain activity of subjects during performance of a task. While this approach has been used extensively for human brain mapping activities, automated classification of subjects based on neural activation patterns is also of interest. However, due to the high dimensionality of the image data, classification accuracy is highly dependent upon the adequacy of the features used in the models. In this work¹, we present a new feature refinement strategy that uses spatial coherence information to eliminate irrelevant features from consideration. For a neurobehavioral disinhibition dataset, we show that this new approach for feature selection using spatially coherent voxels (SCV) outperforms conventional methods.