An important lesson from the first undergraduate statistics course is that the increased sample size leads to a higher power. In brain imaging studies, it is common to combine data collected from multiple study sites to recruit more subjects and increase the reproducibility of scientific discoveries. However, each study site uses MRI scanners from different manufacturers and its own processing protocols, which results in data being corrupted by *unwanted* scanner effects (also termed batch effects). For high-quality data, it is necessary to remove these unwanted scanner effects but, at the same time, preserve biological patterns. The student will develop a data science methodology that removes explicit scanner effects from high-dimensional brain imaging data. A particular application of interest is the cortical thickness data obtained from structural magnetic resonance imaging (MRI). Cortical thickness data reveals an explicit spatial autocorrelation structure, and we hypothesize that the significant source of the scanner effect is heterogeneous spatial autocorrelations by scanners. The student will first conduct exploratory data analysis to visualize and quantify these effects. We will then develop a batch correction method and compare its performance to existing methods using simulation studies. A report summarizing the work is expected by the end of the summer.