Large-scale shape analysis of human brain MRI data

Abstract

Event Abstract Back to Event Large-scale shape analysis of human brain MRI data Arno Klein1*, Joachim Giard2, Forrest S. Bao1, Martin Reuter3, Eliezer Stavsky4, Yrjo Hame4, B N. Nichols5, Satrajit S. Ghosh6 and Jason Tourville7 1 Stony Brook University, Psychiatry & Behavioral Science, United States 2 ICTEAM, Catholique de Louvain, Belgium, Belgium 3 Harvard Medical School, United States 4 Columbia University, United States 5 University of Washington, United States 6 MIT, United States 7 Boston University, United States Human brain shape databases are useful for morphometric studies of healthy and patient populations. They provide scientists with shape measures for comparison with their own MRI data, as well as to train, test, and provide prior information for algorithms that detect, segment, measure, and classify brain structures. Human MRI shape databases have been restricted in the past to measures of labeled region volumes and cortical region thicknesses. These measures are useful for studies of neurogenesis or atrophy in morphological development, degeneration, and disease progression. However, more subtle shape measures may help us to relate structures to behaviors or phenotypes beyond gender, handedness, and relatedness, and have great potential for use in biomarker discovery for clinical diagnosis. As a part of the Mindboggle project, we have created software to extract brain features (labeled regions, sulci, and fundi), and compute shape measures on these features. Currently our shape measures include: mean, Gaussian, maximum, minimum, and principal directions of curvature, travel depth [1], surface area, volume, Laplace-Beltrami spectra [2], and FreeSurfer software-derived measures of depth and thickness [3]. The recent release of our Mindboggle-101 dataset (http://mindboggle.info/data), the largest and most complete set of free, publicly accessible, manually labeled human brain images [4], gives us an unprecedented opportunity to combine automated feature extraction and shape analysis to a large, manually labeled brain MRI dataset. We will present our findings on these shape measures across 101 healthy brains. Acknowledgements This work was funded by the NIMH R01 grant MH084029 (“Mindboggling shape analysis and identification”).