ABSTRACT
The study of brain function with magnetic resonance imaging (MRI), which is sensitive to changes in blood flow and oxygenation [1,2], is a widely used technique, and its applications are growing rapidly-from the early attempts with simple block-designed paradigms to the study of more complex cognitive functions until the study of emotions and behavior [3,4]. Moreover, functional MRI (f MRI) is becoming increasingly important in clinical applications, for example, in neurology and in planning surgical intervention of the brain. The utility of an exploratory data analysis approach is important in order to improve knowledge about the brain function as more complex processes are studied and because it allows the detection and characterization of unexpected phenomena that are not modeled or cannot be modeled
a priori
. Several components may affect signal generation and the experimenter’s model, such as subject movement, physiological changes such as heartbeat and respiration, and noise due to the instrumentation. All these components will bias the results of a model-driven approach that relies on as good a model as possible of the signal as good as possible [5-7]. The knowledge obtained by an explorative approach can be used in confirmatory data analysis (CDA) methods that rely on a precise model of the expected activations. In this framework, exploratory data analysis methods can be seen as hypotheses-generating tools. Moreover, in clinical applications these methods are thought to play an increasingly important role because in these kinds of applications the brain responses typically cannot be modeled in advance. Even if the BOLD signal has been demonstrated to be correlated with the underlying neural activity, several aspects remain to be understood, and exploratory analysis may play a vital role in this. The strength of these exploratory data analysis methods is that information is extracted from the data [8] using only general assumptions, and there is no need of specifying in advance the shape and the extent of a phenomenon. These can be achieved by taking advantage of the multivariate nature of the fMRI data set [9] and the fact that both physiological phenomena of interest, due to the principles of localization and integration of the neural processes [10], and artifacts, may concern measurements in different brain regions. In this chapter we will introduce some methods applied in exploratory data analysis of f MRI data, such as clustering techniques [11-26], principal-component analysis (PCA) [27-32], and independent component analysis (ICA) [33-46]. We will show that even if these methods are powerful tools, in order to improve the knowledge about the brain function the experimenter is required to make some fundamental choices during their applications that can heavily influence the final results.
