A Back Propagation Neural Network for Localizing Abnormal Cortical Regions in FDG PET images in Epileptic Children

In this paper we describe a new method for the definition of cortical abnormalities in epileptic patients. Our objective was to define non-invasively cortical areas with abnormal glucose consumption as measured with positron emission tomography (PET). Using coregistered MRI and PET image volumes, the cortical surface is geometrically parceled into areas of various sizes. The depth of the surface areas is determined using a gray matter mask resulting in small volume elements from which histograms are extracted. Using a back propagation neural network, a system is designed for classifying the histograms into a normal and abnormal group. Those volume elements that are detected as abnormal are marked and the brains are surface rendered in order to allow assessment of cortical abnormalities with respect to cortical landmarks.