Statistical analysis of activation images
Keith J. Worsley · 2003
Abstract Statistical analysis is concerned with making inference about underlying patterns in data that often contain a large amount of random error. This is certainly the case with £MRI data, where the effect of a stimulus may be as little as 1 per cent of the BOLD signal. However, by careful averaging of the data over time, such as averaging the BOLD response at the times when the stimulus is ON, and subtracting the average when the stimulus is OFF, we are often able to detect such a small signal in the presence of considerable background noise. More complex experimental designs require more complex analysis. The type of analysis can be guided by constructing a model for the way in which the BOLD response depends on the stimulus. Such a model must include a component of random error which explains how the observations vary even if the experiment is repeated on the same subject under exactly the same conditions. Statistics can be used to best estimate the parameters in the model, including the variability of the errors. It is this random variability of the errors that can then be used to assess the random variability, or standard error, of the estimated parameters themselves.