This project undertakes the development and application of statistical methodology to neuroimaging. In particular, while brain imaging is a fundamental tool in neuroscience, the statistical treatment of the quantification of such images has lagged behind imaging technology. Numerous statistical problems have not been satisfactorily treated in the analysis of neuroimages. These include: whether and how to normalize positron emission tomography (PET) images on different groups of patients; voxel (volume element) subtraction of images to investigate regions of change in the magnetic resonance Imaging (MRI) scans of the same individuals under different tasks or drugs; multiple comparison issues to safeguard the repeatability of any inference concering a region of apparent activity (where there may be 16,000 voxels per slice); the development of techniques to exploit the spatial correlations of brain imaging as well as the temporal aspects that can occur in repeated scans over time of the same individuals; and the planning of experiments to ensure adequate power. Further, the resolution of these problems is all the more crucial as the imaging technology continues to improve dramatically. Research has been conducted concerning receiver operating characteristic (ROC) methodology that has direct application to the evaluation of different imaging modalities. Work in progress includes methodologies to compare imaging systems as is necessary if, for example, two different PET scanners are used in a study. Collaborative projects have begun on the analysis of data from a PET study involving familial Alzheimer's disease, and an investigation of the variability of metabolites in repeated MR spectroscopic scans (Neuroimaging Branch); and a combined PET~MRI study of induced ischemia in the motor areas of the brains of normal volunteers (Medical Neurology Branch).