Automated identification of human brain features using multi-atlas, registration-based segmentation

Human brain MRI morphometry has largely been restricted to volumes of segmented regions. Features such as sulcus folds, medial surfaces within these folds, and fundus curves that run along the depths of these folds are often used to infer region boundaries. These features may themselves provide morphometric information for use in diagnosing or predicting treatment response for neuropsychiatric disorders [1-5]. There is a significant challenge to doing so, however: sulcus folds are interconnected structures that do not lend themselves to discrete identification, let alone comparison across brains. In this study, we develop and evaluate software for identifying brain features