|
|
Many studies have been conducted using brain imaging data recorded at multiple sites. The environment in which the data are recorded differs from site to site in terms of imaging protocols, preprocessing methods, and characteristics of the subjects. If these different environments affect the data, systematic differences (bias) will occur among the sites, making it difficult to attain valid evaluations during the data analysis, unless measures are taken to deal with the bias during the planning stage. For this reason, in statistical analysis, factors that may cause bias are adjusted as covariates in regression analysis. This paper explains the general linear model and the Bayesian estimation method, which is used in many studies to remove bias (harmonization), with examples. In addition, we discuss the recent remarkable development of machine learning methods for brain imaging data harmonization.