Deepnuparc A Novel Deep Clustering Framework For Finescale Parcellation Of Brain Nuclei Using Diffusion Mri Tractography Haolin He Ce Zhu Le Zhang Yipeng Liu Xiao Xu Yuqian Chen Leo Zekelman Jarrett Rushmore Yogesh Rathi Nikos Makris Lauren J Odonnell Fan Zhang by Haolin He & Ce Zhu & Le Zhang & Yipeng Liu & Xiao Xu & Yuqian Chen & Leo Zekelman & Jarrett Rushmore & Yogesh Rathi & Nikos Makris & Lauren J. O'donnell & Fan Zhang 250307263V2 instant download after payment.

In this work, we propose DeepNuParc, a novel deep clustering pipeline to perform
parcellation of the brain nuclei with dMRI tractography. In our proposed framework,
we first compute a novel voxel-wise connectivity feature representation, with a feature
refinement process using the newly proposed streamline cluster dilation and smoothing.
Next, we design an adaptive k-means-friendly autoencoder framework that can
compress the feature representation and jointly train with the downstream clustering
algorithm. Finally, we achieve fine-scale parcellation of the brain nucleus structure by
clustering voxels into different groups.
A crucial step in our DeepNuParc method involves the explicit reconstruction of
dMRI tractography streamlines to create the feature representation, as opposed to relying
on connectivity probabilities derived from probabilistic tractography [25, 26]. The
explicit reconstruction of tractography streamlines allows us to form streamlines that
pass through a nucleus into streamline clusters, thereby enriching the data representation.
Furthermore, by constructing each voxel’s feature vector based on its traversal by
streamline clusters, we transform a complex high-dimensional brain parcellation problem
into a simpler one-dimensional vector clustering problem. This approach enhances
both the simplicity and robustness of the algorithm.
Our algorithm has good flexibility and compatibility to be modified and customized.