Classification of Cocaine Addicted Patients Using 3D to 1D Hilbert Space-Filling Curve Ordering of fMRI Activation Maps

In analysis of functional magnetic resonance imaging (fMRI), transformation of the 3D brain imaging data to 1D is required for further analyses, which often includes the classification of different groups of participants. The conventional transformation method is linear ordering, which results in a 1D vector that has a high amount of discontinuity which does not preserve the structure of the brain. A Hilbert space-filling curve can better preserve the structure of the brain after the transformation. Features obtained after a transformation based on Hilbert space-filling curve should lead to better classification performance. In this work, we applied Hilbert curve transformation to completely de-identified brain fMRI activation maps from 59 cocaine-addicted and 25 age-matched control participants and classify them as controls vs. patients using machine learning algorithms. Classification based on features from Hilbert space-filling curve ordering resulted in higher classification accuracy of cocaine-addicted patients vs. controls than those of conventional linear ordering.