Predicting individual scores from resting state fMRI using partial least squares regression

An important question in neuroscience is to reveal the relationship between individual performance and brain activity. This could be achieved by applying model regression techniques, in which functional connectivity derived from resting-state functional magnetic resonance imaging (fMRI), is used as a predictor. However, due to the large number of parameters, prediction becomes problematic and regression models cannot be found using the traditional least squares method. We study the ability of fMRI data to predict long-term-memory scores in mild cognitive impairment subjects, using partial least squares regression, which is an adapted method for high-dimensional regression problems. We also study the influence of the sample size on the performance, the stability and the reproducibility of the prediction.