DESCRIPTION (Adapted from Applicant's Abstract): The goal of this application was to achieve the capability to routinely perform automatic, robust, rapid, and accurate multimodality registrations to facilitate comparisons and interpretations of image data set pairs from a variety of sources. The specific aims of the application were to 1) research and develop a quantitative, cost function-based method for automatic registration of multimodality data sets that has utility in clinical and biological research and is based on classical information theory, 2) fully characterize a number of data-related effects on the operation of the algorithm, and 3) demonstrate the algorithm's application to several diverse multimodal data sets in radiology and basic neuroscience where warping is often required for accurate registrations. The registration method proposed is based on the mutual information cost function (MI) which quantifies the mutual information content of two data sources. A geometric mapping that minimizes MI between data sets produces the most spatially correlated, i.e., registered, data sets. An optimizer drives the positions of control points in the homologous data set to effect a mapping between the reference and the homologous data sets that optimizes MI. Since MI is calculated from gray values, it is applicable to iso- and multi-modal sets without the need for preprocessing such as gray level segmentation. The proposed registration method applies to virtually any combination of data sources, both 2D (e.g., autoradiography, electron and light microscopy) and 3D (e.g., CT, MRI, fMRI, PET, SPECT, MEG, and confocal microscopy). Research design consists of several methods to extend our current MI-based prototype to automatically determine the complexity of the registration supported by the data. Three algorithms for automatic control point selection for warping based on local MI will be investigated. Several types of geometric inconsistencies will be systematically studied to elucidate the method's under different conditions. The effect of data set type and transform degrees of freedom on accuracy will be studied by measuring the information content in data from different modalities. Registration accuracy will be validated by use of phantoms; additionally, data sets from rats and humans will be used as a test bed for this work. Analysis of additional data including normal and abnormal cases will demonstrate the efficacy of automated warping registration in the clinical and basic sciences.