Computer-aided visualization and analysis of structural and functional medical images has developed at an extremely rapid pace. Developments in data acquisition, however, have far outpaced developments in automated image analysis and multi-modality integration. In particular, the study of human brain function would be facilitated by a system which automatically integrates structural data with functional data (i.e., structural MRI with functional MRI and evoked potentials), to provide models with high spatial and temporal resolution. We propose to develop a collection of software modules to provide high-performance tissue identification and construction of 3-D models of brain anatomy from structural MRI, which can then be integrated with data from functional MRI and evoked potentials to study human functional neuroanatomy. The system will be trainable by example, providing an easy way to fine-tune performance for specific application domains. During Phase I we will test the feasibility of improving automatic anatomical modeling by using neural network pattern recognition.
