A Novel Path Signature-Based Metric for Quantifying Morphological Characteristics of White Matter Fibers

Motivation: The geometric and morphological properties of human brain fibers can provide more insight into understanding brain mechanisms. However, the morphological analysis of brain white matter (WM) fibers has been relatively scarce. Goal(s): We aim to apply the path signature (PS) to extract more morphological features of brain WM fibers. Approach: A sliding window approach was applied to compute the three-order PSs at each point for every fiber, resulting in 75 PS maps for each brain. The gender classification models were constructed based on these maps and fractional anisotropy (FA). Results: The accuracies of all classification models trained on PS maps consistently outperform FA. Impact: The results demonstrate that PS features are more sensitive in capturing morphological differences between male and female WM fibers than FA. PS features can be used to characterize the multi-dimensional morphological features of brain WM fibers.