13 NovSfN2014 Poster

Exploring neuroanatomical variation in a large database of manually-labeled human MRI scans

Andrew Worth¹, Jason Tourville²

¹ Neuromorphometrics, Inc., Somerville, MA
² Dept. of Speech, Language, and Hearing Sciences, Boston University, Boston, MA

Abstract (download poster, 11.7MB)

In automated labeling of neuroanatomy, it is important to know both what is normal and what is normally different. In this work, we describe methods to understand, represent, and quantify variation of human brain anatomy using 114 consistently labeled scans. Seventy-two regions were manually labeled to comprehensively cover the brain including: cerebral and cerebellar gray and white matter, ventricles, brain stem, accumbens, amygdala, caudate, hippocampus, pallidum, putamen, and thalamus according to our “General Segmentation” protocol, along with 51 parcellation units based on 36 sulci according to the BrainColor cortical parcellation protocol. To begin to explore variation, it is necessary to be able to see anatomy individually and in aggregate. Visual inspection of slice images, 3D surfaces, and probabilistic renderings using color and transparency leads to understanding of shapes and spatial relationships. Statistics calculated on volume, shape, overlap, and position provide a quantitative characterization. For example, a probabilistic Dice coefficient is used to compute how well an individual region matches its probabilistic average. High-dimensional non-linear mapping can be used to eliminate variation. Alternatively, in this work probabilistic atlases were created in two ways: 1) using a thin plate spline warp to put individual scans into Talairach space (defined by anterior and posterior commissure, the midline and extents in 3 dimensions), and 2) using standard algorithms to transform scans into MNI Space. We compare individual scans to these atlases and the atlases to each other for all scans and also for atlases created from 3 age groups: child (5-15 years old), adult (18-38), and older (40-96). A “dynamic” atlas was also created by averaging subjects near ages ranging from youngest to oldest. To present a feel for the range of variation in our sample, the best and worst matching individual anatomy is shown overlaid onto atlases, and atlases are compared to each other visually and statistically. We conclude by discussing two general methods of representing anatomical variation, 1) Topological: a case-by-case definition of commonly found configurations with percentages observed such as that described by Ono (Thieme Medical, New York, 1990), and 2) Topographical: probabilistic atlases show bright and high-contrast anatomy for less variable regions, and fuzzy dark areas for more variable anatomy. We briefly illustrate some uses of the knowledge of brain variation for clinical and neuroscientific applications involving transcranial magnetic stimulation and diffuse optical tomography.