ABSTRACT
Functional Magnetic Resonance Imaging (fMRI) is a tool for studying brain function, i.e., which neural systems are the brain substrate for a particular behavior. fMRI is noninvasive, meaning that neither surgery nor ionizing radiation are used to generate the images. It can also localize activity to a few millimeters. For these reasons, fMRI as a research area has exploded over the last 10 years. In task-based analysis,2 the brain is treated as a black box to which a stimulus is applied and from which a response is measured. The fMRI measurement itself is related to the amount of deoxygenated hemoglobin in the blood and so it is called the Blood Oxygen Level Dependent (BOLD) signal. While dependent upon many other factors, the BOLD signal gives an indication of how much blood is flowing to a particular location in the brain at a particular time. Behavioral and thought processes cause neurons in a small area to become active, which increases blood flow to that area to supply the metabolic demand of the neural activity. The more intense the neural activation, the larger the increase in blood flow will be (though the relationship is not necessarily linear or straightforward). The analysis strategy is to correlate the BOLD waveform with the known time course of stimulation. Brain areas engaged in the task will have a BOLD signal that is correlated with the task; brain areas that are not engaged in the task will be uncorrelated. This allows the characterization of the neural substrate of behavior in terms of the locations in the brain that are engaged, the amount of brain tissue recruited, and the strength or amplitude of the response. This may be clinically relevant for an individual person for diagnosis, for treatment planning, for monitoring treatment response, or for assessing disease course. We are also interested in how populations vary, e.g., how do the location, intensity, and size of neural activation change between those diagnosed with schizophrenia versus those who are deemed clinically healthy? This leads to potentially four levels of analysis: (a) time series analysis of an individual at a particular visit date, (b) longitudinal analysis (i.e., from visit to visit), (c) between subjects within a defined population, and (d) between defined populations. When studying large samples, it is often necessary to scan them at different study locations, which can add a fifth analysis level (site). Each of these levels introduces a set of noise that will increase variability and/or introduce bias to the final result. This chapter describes how fMRI is analyzed at each of these levels, the sources of noise at each level, and ways to control the noise. We start with a case study to familiarize the reader with the terminology of fMRI as well as provide motivation for conducting a study. Next, we describe the analysis at the first and higher
levels based on the general linear model (GLM). The biophysics of fMRI are then summarized to introduce the reader to the physics of the measurement and a foundation for understanding the sources of the noise. Finally, the sources of noise at the various levels are surveyed along with methods to reduce their impact.
