Accurate non-iterative modelling and inference of longitudinal neuroimaging data

In recent years, increasing efforts have been made to collect longitudinal neuroimaging data in order to study how brains change over time. However, the popular methods used to analyse such kind of data may not always be appropriate (e.g., overly sensitive to model misspecifications, difficult to specify adequately or prohibitively slow to compute) and may sometimes lead to erroneous conclusions. Motivated by these shortcomings, in this dissertation, we have proposed and studied the use of an alternative method, referred to as “the Sandwich Estimator method”, and have demonstrated that it is a fast, easy-to-specify and accurate option to analyse longitudinal or repeated-measures neuroimaging data.