Meeting Banner
Abstract #0778

Fast and Fully Automated Clustering of Whole Brain Tractography Results Using Shape-Space Analysis

Greg D. Parker1, David Marshall2, Paul L. Rosin2, Nicholas Drage3, Stephen Richmond3, Derek K. Jones1

1CUBRIC, School of Psychology, Cardiff University, Cardiff, South Glamorgan, United Kingdom; 2School of Computer Science, Cardiff University, Cardiff, South Glamorgan, United Kingdom; 3School of Dentistry, Cardiff University, Cardiff, South Glamorgan, United Kingdom

We propose a novel method for fully automated segmentation of large tractography datasets. By measuring the modes and magnitudes of streamline shape variation within the brain, we are able to build a white matter shape space in which streamlines belonging to particular anatomical features consistently project to distinct sub-regions; thus allowing us to segment unseen streamline data by observing their projected positions. An additional advantage of this technique is the computationally trivial nature of the projection process which, when compared to other techniques with similar aims, significantly reduces both run time and memory footprint.

Abstract PDF Presentation Video

Keywords

according accounted acquisition affine anisotropy applied appropriate approximately assigned automated automatically axes axis bank belonging blue brain brains bundles casting classified cluster clustering clusters comparatively component computational computationally computer consistent contents coordinate corpus correction covers creation cyan damped dataset datasets dell dentistry derived describe descriptive descriptor descriptors detail deviation dimensions distinct eigenvalues eigenvectors eliminates embedding examination expected expensive fast feature form fractional fully healthy human implied individual individually initial inter isotropic kingdom knots label learn load manual matrices motion native nearest need noise novel operation origin parker partitioning plot practical principal problem project projected projection proximity purple qualitatively rapid rather reached recognition reduce reduces refer registered remain removes resolution respectively robust rotations school science segmentation segmentations segmenting separated shape shaped shapes similarly simple solutions south space spatial still streamline streamlines structures subject subset sufficient sufficiently support template though thresholds tracking tracts train training trajectories transformations trivial variation variations vector white whole