• general linear model
• fMRI
• exploratory data analysis
• bayesian inference
• hierarchical linear model
• independent component analysis

Functional Magnetic Resonance Imaging (fMRI) is a technique that allows the study of sensory, motor, and cognitive functions of the brain.
This thesis deals with the development of methods for the analysis of fMRI data: two major aspects of data analysis represented by confirmatory, or hypothesis driven, and exploratory, or data driven, approaches are taken into account.
Within the hypothesis driven approaches, two mixed effects models were evaluated: the first uses an expectation maximization algorithm, for the simultaneous estimation of first and second level parameters of the model. The second employs a two stage procedure for estimating separately subjects and group related parameters. The two methods gave similar results on simulated dataset, while the first showed a better behaviour in real data set analysis we presented.
Data driven methods as independent component analysis (ICA) can be used to solve the drawbacks of hypotheses driven methods, as the impossibility of detecting unmodelled or unexpected phenomena.
The effectiveness of ICA in removing movement related artefacts, temporally correlated with the phenomenon of interest, has been evaluated, showing the strong influence of image noise on the final results.
A novel approach to classify spatially Independent Components, and overcoming model order indeterminacy was proposed: the method here introduced performs a hierarchical clustering of the components by using a similarity measure derived from mutual information.
This algorithm tested on simulated as well as real fMRI datasets has been proved to be a valid tool to detect and merge components derived from the splitting process due to overestimation of model order.