Neural Registration and Segmentation of White Matter Tracts in Multi-modal Brain MRI

Noa Barzilay, Ilya Nelkenbaum, Eli Konen, Nahum Kiryati, Arnaldo Mayer

Abstract: Pre-surgical mapping of white matter (WM) tracts requires specific neuroanatomical knowledge and a significant amount of time. Currently, pre-surgical tractography workflows rely on classical registration tools that prospectively align the multiple brain MRI modalities required for the task. Brain lesions and patient motion may challenge the robustness and accuracy of these tool, eventually requiring additional manual intervention. We present a novel neural workflow for 3-D registration and segmentation of WM tracts in multiple brain MRI sequences. The method is applied to pairs of T1-weighted (T1w) and directionally encoded color (DEC) maps. Validation is provided on two different datasets, the Human Connectome Project (HCP) dataset, and a real pre-surgical dataset. The proposed method outperforms the state-of-the-art TractSeg and AGYnet algorithms on both datasets, quantitatively and qualitatively, suggesting its applicability to automatic WM tract mapping in neuro-surgical MRI.