Characterization of the univariate and multivariate techniques on the analysis of simulated and fMRI datasets with visual task

Current analytical techniques applied to functional MRI (fMRI) data may be generally divided into two parts: univariate and multivariate techniques. It is therefore our attempt to evaluate and inter-compare their respective algorithms on simulated and fMRI visual task data sets. In this study, the two representative univariate approaches, including the correlation and the specified-resolution wavelet analytical methods, and three multivariate based independent component analysis (ICA) approaches; including the Infomax ICA, the Fast ICA, and the JADE ICA are used for the purposes. Two simulated spatial sources with different time courses and noise levels and one fMRI dataset with visual task were employed for inter-comparisons. Strategies for quantifying the performance of these techniques, the correlation analysis and receiver operating characteristics (ROC) are used to evaluate their respective accuracies on estimated time-courses and spatial layouts from the simulated and the fMRI visual task dataset In our results, it demonstrates that the multivariate techniques generally outperformed the univariate techniques, among which the Fast ICA performs satisfactory well on temporal and spatial accuracy.