Data-Augmented Deep Learning for Tractography Parcellation of Diffusion MRI with Incomplete Field of View

Motivation: Tractography parcellation is crucial for quantitative analysis of white matter connectivity and microstructure related to brain health and disease. However, real-life dMRI scans often have incomplete Field of View (FOV), and robust tractography parcellation methods do not yet exist. Goal(s): We aim to propose a robust framework for parcellation of tractograms with inferior FOV cutoff. Approach: We propose a deep learning framework for effective tractogram parcellation with incomplete FOV by leveraging data augmentation with synthetically cut tractograms. Results: Evaluations on datasets with simulated and real-world FOV limitations indicate that our method enables effective parcellation of tractography affected by incomplete FOV. Impact: dMRI scans with incomplete FOV are common in real-life clinical and large-scale research datasets and pose great challenges to identification of anatomical tracts. We propose the first deep learning framework to achieve robust parcellation of tractography affected by incomplete FOV.