Structured and automated analysis techniques for morphometric mapping of the human brain

The analysis of brain morphology in the field of neuroimaging has become a complex and arduous task. Many steps, automated, semi-automated, and manual, are required to go from raw data to discernable results. The required tasks are not always reliable in their automated results and manual approaches are time consuming and subject to their own reliability questions. Once completed, important study factors and unforeseen results may be missed due to limited power and statistical design. A novel approach to study large populations for anatomic differences in brain structure requires reliable automated methods, an analytic and statistical design that provides for appropriate data mining and the presentation of correlations between independent variables and interesting morphologic vectors, and a computational environment that allows these tasks to be achieved and derived in a timely manner. A meta-algorithm uses the application of multiple task-oriented algorithms simultaneously and combines their results into a single improved result. The development of automated and trainable meta-algorithms for specific tasks in neuroimaging provides for more robust and improved results over a variety of input data. This allows the analysis of large diverse populations in a reliable and unsupervised framework. A novel meta-algorithm, the brain extraction meta-algorithm, is provided with four separate brain extraction algorithms and a spatial registration algorithm in a mathematical context for combining their results. It achieves more robust and reliable results than any of the contributing algorithms achieve on their own. Using non-linear spatial normalizations of a population of brains to an internally defined atlas space and the tensor analysis of the differences between normalization fields along a proper manifold allows for the discovery of anatomic differences occurring across the population. Analyzing the millions of derived variables in a multivariate statistical context with the clustering of results allows the presentation of anatomically differentiated groups in the population and the analysis of the morphologic differences between those groups. Intuitive construction and achieving these computationally expensive procedures within a reasonable timeframe is provided by a novel distributed processing and parallel execution visual programming environment, the LONI Pipeline Processing Environment.