TractNet: A Deep Learning Approach on 3D Curves for Segmenting White Matter Fibre Bundles

Fibre tractography techniques generate a large number of fibre streamlines that are densely distributed throughout the brain. These 3D fibre curves are classified into a discrete collection of white matter bundles for tract-based quantitative analysis which is a modern effective methodology for clinical demands in brain development, aging and disease. Although several approaches have been developed to segment the 3D fibre curves into anatomically meaningful fibre bundles, it is still a challenging task due to the complex and large volume of 3D curve data representation. In this paper, we propose a deep learning architecture (called TractNet) to segment ten major fibre bundles and to segment the curves which do not belong to these ten major bundles. Proposed architecture consumes 3D fibre curves in their raw data format. Moreover, proposed architecture has two channel attention modules to boost the segmentation performance. We demonstrated quantitative and the visual evidence of significant performance of the proposed model. The TractNet showed strong performance that is comparable to or better than the state-of- the-art in terms of experimental evidence.