Leveraging Longitudinal Data to Improve BrainChart Calibration for Small Study Sample Sizes

Longitudinal brain imaging studies offer valuable insight into trajectories of brain structures over time; however, changes in acquisition scanners and protocols can introduce biases in the resulting measures. Although BrainChart provides a framework for calibrating FreeSurfer-derived brain structure measurements between studies or time points, population samples with small subject numbers ≤ 100 are known to give unstable sample effect parameter estimates. Under the assumption that centiles of control subjects are stable across time points, we present a method to improve reliability of population sample effect parameter estimation for time points in longitudinal studies that have small numbers of control participants but include a nested sample with repeat scans of some control participants at multiple time points. We verify the accuracy of this method using both simulated and real datasets, which demonstrate comparable estimates close to the ground truth and improved confidence intervals at smaller sample sizes than the original method. Additionally, our approach confirms accuracy across both time points and scanners. The software, which is available on GitHub, offers a means to extend the value of brain imaging data acquired using variable scanners and/or protocols for longitudinal studies, thus maximizing the value of brain imaging data in established cohorts.