Neural networks for fMRI spatio-temporal analysis

Most of the analysis techniques applied to functional magnetic resonance imaging (fMRI) consider only the temporal information of the data. In this paper, a new method combining temporal and spatial information is proposed for the fMRI data analysis. The nonlinear autoregressive with exogenous inputs (NARX) model realized by radial basis function (RBF) neural network is used to model the fMRI data. This new approach models the fMRI waveform in each voxel as a regression model that combines the time series of neighboring voxels together with its own. Both simulated as well as real fMRI data were tested using the proposed algorithm. Results show that this new approach can model the fMRI data very well and as a result, can detect the activated areas of human brain successfully and accurately.