INTRODUCTION TO STATISTICAL ANALYSIS OF BOLD DATA

The statistical analysis of Blood Oxygenation Level Dependent (BOLD) data is a critical part of brain mapping with functional magnetic resonance imaging (fMRI). Many creative statistical methods have been proposed, and there has been considerable debate about which is the “correct” approach. Given the flexibility of fMRI and the range of experiments that is possible, it seems likely that a number of different statistical processing approaches can be applied to yield useful data. Indeed, a pluralistic analysis strategy applying several methods to the same data may be the best approach for pulling out and evaluating the full information content of the fMRI data (Lange et al., 1999). The goal of this chapter is to highlight some of the important aspects of statistical thinking about BOLD data analysis, rather than to provide a comprehensive review of different approaches. We will focus on the general linear model, which encompasses many of the techniques commonly used (Boynton et al., 1996; Friston et al., 1995; Friston, Jezzard, and Turner, 1994; Worsley et al., 1997). In the first section, we introduce the need for a statistical analysis and some of the basic ideas and strategies. In the second section, the general linear model is considered in more detail, emphasizing the geometrical view of the analysis. Chapter 19 focuses on how the statistical analysis sheds light on how to design an efficient experiment and provides a framework for comparing the relative merits of blocked and event-related stimulus paradigms.

Separating True Activations from Noise

The magnetic resonance (MR) signal change during activation due to the BOLD effect is quite small, on the order of 1% for a 50% change in cerebral blood flow (CBF) at 1.5 T.