Automatic classification of sulcal regions of the human brain cortex using pattern recognition Kirsten J. Behnke a,b , Maryam E. Rettmann a,b , Dzung L. Pham c , Dinggang Shen d , Susan M. Resnick b , Christos Davatzikos d , Jerry L. Prince *e a Dept. of Biomedical Engin., Johns Hopkins Univ., 3400 N. Charles St., Baltimore, MD 21218 b Gerontology Research Center, NIA/NIH, 5600 Nathon Shock Dr., Baltimore, MD 21224 c Neuroradiology Div., Johns Hopkins Univ., 600 N. Wolfe St. Baltimore, MD 21287 d Sect. of Biomed. Image Analysis, Dept of Radiology, Univ. of Penn., Philadelphia, PA 19104 e Dept. of Electrical Engin., Johns Hopkins Univ., 3400 N. Charles St., Baltimore, MD 21218 ABSTRACT Parcellation of the cortex has received a great deal of attention in magnetic resonance (MR) image analysis, but its usefulness has been limited by time-consuming algorithms that require manual labeling. An automatic labeling scheme is necessary to accurately and consistently parcellate a large number of brains. The large variation of cortical folding patterns makes automatic labeling a challenging problem, which cannot be solved by deformable atlas registration alone. In this work, an automated classification scheme that consists of a mix of both atlas driven and data driven methods is proposed to label the sulcal regions, which are defined as the gray matter regions of the cortical surface surrounding each sulcus. The premise for this algorithm is that sulcal regions can be classified according to the pattern of anatomical features (e.g. supramarginal gyrus, cuneus, etc.) associated with each region. Using a nearest-neighbor approach, a sulcal region is classified as being in the same class as the sulcus from a set of training data which has the nearest pattern of anatomical features. Using just one subject as training data, the algorithm correctly labeled 83% of the regions that make up the main sulci of the cortex. Keywords: sulcal labeling, human brain cortex, pattern recognition, deformable models, sulci, atlas 1. INTRODUCTION The surface of the human brain cortex is made up of many convoluted folds separated by spaces known as sulci. The classification of these sulci is an important step in many neuroimaging studies, which seek to analyze morphological changes in regions of interest on the cortex that are typically defined by the primary sulci (cf. 1-5 ). A sulcal classification scheme would facilitate a parcellation of the cortex into regions that are both functionally and anatomically important. In this work, we present a method to classify the key sulci with the future goal of parcellation in mind. There are many software programs available that a trained neuroanatomist or technician could use to manually label sulci on the brain. 6-8 Unfortunately, this task is both difficult and time-consuming, and thus a scheme for automatic labeling is necessary. Past efforts at an automated labeling algorithm have involved warping a prelabeled atlas to a preprocessed image of a test brain (cf. 9-13 ), thereby transferring labels from the deformed atlas to the appropriate locations on the test brain. Other more recent efforts have favored supervised algorithms in which sulci are matched with models from a training database based on characteristics such as shape, location, or structure. 14-18 Our method combines key concepts from both approaches. The challenge facing all of the approaches is the high variability in both shape and structure of sulci from subject to subject. To address this problem, our algorithm begins with a data driven sulcal segmentation that is unique to the subject, followed by a classifier which identifies the segmented sulci based on features extracted using a deformable atlas. In this work, we present an algorithm for automatically classifying sulcal regions of the human cortex. We begin with a definition of sulcal regions and a review of the work we have done previously in segmenting these * Send correspondence to prince@jhu.edu; phone: 1 410 516-5192; fax 1 410 516-5566; http://iacl.ece.jhu.edu