A Framework to Compare Tractography Algorithms Based on Their Performance in Predicting Functional Networks

摘要

Understanding the link between brain function and structure is of paramount importance in neuroimaging and psychology. In practice, inaccuracies in recovering brain networks may confound neuro-physiological factors and reduce the sensitivity in detecting statistically robust links. Hence, reproducibility and inter-subject variability of tractography approaches is currently under extensive investigation. However, a reproducible network is not necessarily more accurate. Here, we build a statistical framework to compare the performance of local and global tractograpy in predicting functional brain networks. We use a model selection framework based on sparse canonical correlation analysis and an appropriate metric to evaluate the similarity between the predicted and the observed functional networks. We demonstrate compelling evidence that global tractography outperforms local tractography in a cohort of healthy adults.