Autism is a complex disorder of early onset, involving odd and repetitive movements, severe social disability, deficits in social cognition, and disruption of language. While the multiple signs and symptoms in autism suggest several different brain systems are likely involved in its pathobiology, it remains the fact that most efforts aimed at the analysis of neuroanatomical structure related to autism from (primarily Magnetic Resonance (MR) images have been limited to the measurement of rather gross features, such as overall brain size and cross sectional area, or measurements of the corpus callosum and cerebellar vermis, using fairly small samples. These limitations are in large part because, to date, manual and computer-assisted, semi-automated segmentation/measurement of neuroanatomy is a tedious, labor-intensive, and costly process, subject to human variability. The research proposed here is aimed at the further development of an image analysis strategy that will accurately, reproducibly, robustly and efficiently analyze neuroanatomical structure relevant to autism from 3D high resolution MR images. At the core of this effort are unique mathematical approaches to: i.) segment cortical structure using coupled differential equations to simultaneously locate the gray/white and gray/CSF surfaces; ii.) segment subcortical structure by adding shape and inter-structure spatial relationship priors to an approach that integrates boundary finding and region growing; and iii.) nonlinearly register regional neuroanatomical structure to create atlases and match them to segmented information for the purpose of labeling cortical gyri and guiding the subcortical segmentation process. A key feature of the approach is that the final labeling and measurement that is performed is done by carefully focusing on individual regions of the brain, one at a time. The accuracy and robustness of the individual algorithm components to imaging parameters, field inhomogeneities and noise will be demonstrated by validating segmentation, registration, labeling and measurement algorithm results from synthetic data created using an MR image simulator against gold standard source images. The utility of the image analysis strategy for deriving robust, accurate measures in a variety of cortical and subcortical brain regions relevant to autism will be evaluated by running the algorithm on a cohort of 30 normal control and 30 subjects having autism and/or related conditions, sampled from a large, well characterized and separately NIH-funded subject database.
Showing the most recent 10 out of 36 publications