Subject-specific estimation of missing cortical thickness maps in developing infant brains

To accurately chart the dynamic brain developmental trajectories in infants, many longitudinal neuroimaging studies prefer having a complete dataset. Unfortunately, missing data at certain time points are unavoidable in longitudinal datasets. To better use incomplete longitudinal data, we propose a novel method to estimate the subject-specific vertex-wise cortical thickness maps at missing time points, by using a customized regression forest, Dynamically-Assembled Regression Forest (DARF). DARF ensures spatial smoothness of the estimated cortical thickness maps and also the computational efficiency. The proposed method can fully exploit the available information from the subjects both with and without missing scans. Our method has been applied to estimate the missing cortical thickness maps in a longitudinal infant dataset, which includes 31 healthy subjects, with each having up to 5 scans. The experimental results indicate that our method can accurately estimate missing cortical thickness maps, with the average vertex-wise error less than 0.23 mm.