The specific aim of this proposal is to automatically identify/match brain features based on a geometric and parametric analysis of their shapes, by means of a Bayesian framework derived from the face recognition literature. Mindboggle, a freely available software package for performing automated anatomical brain labeling, will serve as the software infrastructure for implementing the Bayesian framework. The secondary aim is to further develop Mindboggle to automatically label an entire brain based on these probabilistic matches. These anatomical labels provide a consistent, convenient, and meaningful way to communicate, classify, and analyze biomedical research data. Clinical research applications of labeled regions include volumetric and shape analysis of brain regions over time or across conditions, and region-specific analysis of, for example, fMRI or PET activity data. There are two longterm research objectives: (1) establish a consistent, automatic, and fast method for labeling brains with an accuracy and precision comparable to that of manual labelers, and (2) obtain shape characteristics of anatomical regions, their variations and covariations in healthy subjects and patients, and their relationships to microstructure, connectivity, physiology, and functional activity for genetic, behavioral, developmental, and clinical research. Brain structures will be extracted from human brain MR image data and analyzed (described and compared) using geometrical and parametric approaches, for the purpose of identifying the brain structures to which the shapes correspond and characterizing their morphological variability across brains. Mindboggle algorithms for fragmenting these skeletons. Geometric analysis of shapes will include the gross shape descriptors such as mean distance between two coregistered shapes, their volumes, degree of overlap, etc. Parametric analysis will employ quantitative shape discrepancy metrics derived by an active surfaces model. These measures of similarity between shapes will be applied to a large dataset of manually labeled brain data and incorporated in a Bayesian framework for the purpose of estimating the probability of a given shape corresponding to a particular brain structure. Image processing will entail skeletonization of brain cortical folds combined with revised Additional contributions will be a dataset of manually labeled brains for research and educational purposes, and a database of individual brain morphological variability derived from this dataset.
Automatically characterizing the shapes of brain structures and labeling the anatomy of brain image data in an accurate, fast, and consistent manner is of immense value to clinical researchers interested in the application of brain imaging to mental health. Anatomical labels provide a consistent, convenient, and meaningful way to communicate, classify, and analyze biomedical research data. Clinical research applications include volume and shape analysis of brain regions over time, across conditions, and across groups of patients or healthy subjects, as well as analysis of fMRI or PET activity data acquired from these regions.
| Ghosh, Satrajit S; Klein, Arno; Avants, Brian et al. (2012) Learning from open source software projects to improve scientific review. Front Comput Neurosci 6:18 |
| Bao, Forrest Sheng; Giard, Joachim; Tourville, Jason et al. (2012) Automated extraction of nested sulcus features from human brain MRI data. Conf Proc IEEE Eng Med Biol Soc 2012:4429-33 |
| Klein, Arno; Ghosh, Satrajit S; Avants, Brian et al. (2010) Evaluation of volume-based and surface-based brain image registration methods. Neuroimage 51:214-20 |
| Klein, Arno; Andersson, Jesper; Ardekani, Babak A et al. (2009) Evaluation of 14 nonlinear deformation algorithms applied to human brain MRI registration. Neuroimage 46:786-802 |
