Subspace identification methods and fMRI analysis

The main goal of this paper is to propose application of modern multidimensional systems identification algorithms of the subspace identification theory in the context of fMRI data analysis. The methods originated in 1990s in the field of process control and identification and yield robust linear model parameter estimates for systems with many inputs, outputs and states. Our ultimate goal is to establish an alternative to the DCM analysis procedure which would eliminate its main drawbacks, namely the need to pre-define the models structure. The paper discusses results based on simulated data provided by the DCM simulator in the SPM toolbox. Several scenarios are presented, with varying amount of noise and number of data samples.